Spacetime Folding

The maze becomes a path.

The combinatorial chasm

The Traveling Salesman Problem is the canonical combinatorial wall. Visit every city once, return home, keep the path short. The decision variant is NP-complete. Exact search scales exponentially with the number of cities n. Enumeration of all n! tours becomes prohibitive long before n exceeds 20. Twenty cities drown a brute-force machine. Real logistics, circuit routing, and molecular conformation need dozens or hundreds — and they need the answer while the trucks are still loading.

Nearest-neighbor is fast and typically 10–20 percent worse than optimal. Simulated annealing can climb, but it wants tuning and time. The gap is a method that refuses brute force, uses the structure that is already there, and still hands back a high-quality tour. That is why spacetime folding exists. The maze is combinatorial. The exit is geometric.

Octonions as the phase algebra

The algebra is the octonions: eight dimensions, seven imaginary units, multiplication read from the Fano plane. Two units multiply to a third along a line of that plane. Reverse the order and the sign flips. The rules are alternative and not associative — they stay rich enough to hold high-dimensional geometry without collapsing into a smaller algebra.

Those octonions sit against the E8 lattice. The house already addresses the archive on 240 E8 roots. The same exceptional lattice is the phase space. The 240 roots generate 480 distinct multiplication triples — each triple unique under the Fano rules. That is not decoration. It is the multiplication table that governs how phases align. Octonions are the phase algebra because their product is the dynamics: a Kuramoto oscillator network whose coupling lives on that table.

• Fano plane — the mnemonic for seven imaginary units and their cyclic products.
• 240 E8 roots — the live address set of the lattice, and the neighborhood of local moves.
• 480 multiplication triples — distinct products on those roots.

Unfold, resonate, fold

Three phases. Unfold embeds city coordinates into E8 so Euclidean distance becomes a coupling in eight dimensions. A cuboctahedron — the vector equilibrium — seeds a uniform first phase so the search does not start leaning. Resonate runs a Kuramoto-style oscillator on every city: each phase feels every other, with nearby cities pulling harder. Fold sorts the locked phases and hands back the tour.

That pipeline is the engine sentence: Unfold → Resonate → Fold. It is not a random walk through permutations. The embedding is the map. Synchronization is the work. Extraction is a sort. Distances build in O(n²), then O(1) lattice snaps along root directions. The fold is an ascending phase sort — O(n log n) — then a short 2-opt polish that clears crossings.

Unfold embeds the cities. Resonate locks the phases. Fold hands back the tour.

Geometry solves combinatorics

The load-bearing claim is physical. Cities become oscillators. Inverse-distance coupling makes neighbors want the same stretch of tour. The natural frequencies come from the E8 embedding, so the lattice is a prior, not a costume. Too little coupling and nothing syncs. Too much and every phase collapses to one useless note. A golden-section search lands the coupling near K* ≈ 2.663. Convergence is O(log n) coupling cycles — about seven cycles at a hundred cities, about ten at a thousand.

Self-organization is the point. The system does not enumerate tours. It falls into a low-energy phase order, and that order is a good route. Physics finds the path. A later 2-opt pass is polish, not the search. A combinatorial wall turns into a synchronization problem the geometry already knows how to finish.

Geometry solves combinatorics. The tour emerges from phase, not from enumeration.

What the paper measured

Headline numbers for n = 50: about 9.5 percent better than nearest-neighbor, and a 43 ms GPU runtime in that write-up. A hundred-city write-up stays under 200 ms. Four cities come back optimal in 2.3 ms. Those are paper results on the instances they ran. They are not a live service-level agreement.

Nearest-neighbor is the baseline because it is what you get when you are greedy and in a hurry. Nine and a half percent shorter than that greedy tour, in tens of milliseconds, is the maze becoming a path. Brute force at fifty cities is not a competitor. It is a fossil. The method is inherently parallel: every oscillator updates at once, which is why a GPU write-up can stay in the tens of milliseconds while factorial search leaves the building.

• n = 50 — 9.5% better than nearest-neighbor; 43 ms in the paper GPU write-up.
• n = 100 — under 200 ms in the same paper benches.
• n = 4 — optimal tour in 2.3 ms on the small case.

The door after Memory

Memory is the house writing a fact and reading the same fact back. This door asks the same geometry for a harder job. Cities unfold into the lattice. Oscillators resonate until the phases lock. The fold is a tour. Combinatorial search becomes a path you can dispatch. A buyer who already wants the archive wants this next: the address space that orders a file now ordering a wall of cities. Unfold → Resonate → Fold is the sentence. The paper’s n = 50 write-up is the want-number — 9.5 percent better than nearest-neighbor, 43 milliseconds on that GPU.

On the public catalog this remains a research door, never a SKU. P=NP proven is forbidden. A fast fifty-city tour is a geometric heuristic with paper numbers; it does not decide the millennium problem. Space Fold runs at 1 Hz when it runs at all, and the live state is DORMANT until the door is lit. If clocks appear, they stay paired: 13.834 µs isolated, 5,127 µs under load — memory clocks, not a fold hop. $20 million is an ask to land the archive first.

Next in the catalog: NeoDrive.