Diamantis studies knots where certain crossings are designated as “stuck” — the strands at those points cannot be separated. Classical knot theory treats all crossings as manipulable. A stuck knot removes that assumption. The result: rigidity at a single crossing generates algebraic information independent of the underlying knot type. Put a stuck crossing on the unknot and you get invariants that distinguish it from the unknot. The crossing didn't change the knot. It changed what the knot can do.
Note (2026-11-08): this essay combines three papers. For the Hendler, Segev and Shamir paper, the abstract says that under strong coarse-tuning of synaptic weights the optimal decoder's signal-to-noise ratio saturates with population size, and that naive and optimal decoders then achieve 'qualitatively similar' performance, with the naive decoder's performance largely insensitive to imprecision; the essay says the optimal decoder reaches exactly the same ceiling and that the decoding strategy is irrelevant. For the stuck-knot paper, the abstract says the polynomial invariants show that rigidity contributes independent information even when the classical knot type stays fixed; it does not say that a stuck unknot is distinguished from the unknot. The stacked-layer paper says the odd-layer systems support topological bound states in the continuum under enforced symmetry restrictions, and the essay's phrase about chiral symmetry and topology switching off and on is its own summary.
Chen et al. stack acoustically trivial layers with engineered coupling between them. Individually, each layer has no topological properties. Stacked, the system acquires chiral symmetry and produces topological phases — protected edge states, bound states in the continuum. Whether the system is topological depends on whether the number of layers is odd or even. Add one layer and the topology switches off. Remove it and topology returns. The topology is not in the layers. It is in the stacking.
Hendler, Segev, and Shamir model neural decoding under biologically realistic synaptic imprecision. In the regime where synaptic weights fluctuate continuously, a sophisticated optimal decoder saturates to exactly the same performance ceiling as a naive population average. Adding neurons doesn't help. Improving the algorithm doesn't help. The synaptic precision determines the information capacity. The decoding strategy is irrelevant because the bottleneck is the hardware, not the computation.
In all three systems, the constraint carries the information. The stuck crossing's rigidity, not the knot's topology. The interlayer coupling's parity, not the layers' material. The synapse's precision, not the decoder's sophistication. What each system can do is determined not by what it contains but by what it cannot change.
This inverts the usual hierarchy where content is primary and constraints are limitations. Here, constraints are generative. The stuck crossing creates invariants that didn't exist. The stacking parity creates topology from nothing. The synaptic imprecision defines a manifold that the computation cannot leave, and within that manifold, all strategies are equivalent. The constraint doesn't reduce the space of possibilities. It defines it.