Concrete example first

Let (H_t={\bot<m<\top}). Define

  • (\sigma_t(\bot)=\bot),
  • (\sigma_t(m)=\top),
  • (\sigma_t(\top)=\top).

Applying (\sigma_t) once pushes uncertain recognition (m) to a stable value (\top). Applying it again changes nothing. That is stabilization.

Formal definition

For each trace (t), a stabilizer is an endomorphism (\sigma_t:H_t\to H_t) that is:

  • monotone: (a\le b \Rightarrow \sigma_t(a)\le \sigma_t(b)),
  • idempotent: (\sigma_t(\sigma_t(a))=\sigma_t(a)).

In GFRTU, stabilizer maps are part of the primitive recognition dynamics and must be compatible with reindexing across traces.

Why it matters in GFRTU

  • It models convergence to stable recognitions.
  • It defines one half of the fixed-layer condition (\sigma_t(a)=a).
  • Together with drift, it determines the stable core that survives dynamics.