geometry

L = sum of boundary areas. the foam lives in U(d). cells are Voronoi regions of the basis matrices under the bi-invariant metric; boundaries are geodesic equidistant surfaces; Area_g is the (d^2 - 1)-dimensional Hausdorff measure. bases in general position tile non-periodically.

the Voronoi tiling is a realization choice: it determines adjacency (which pairs share a boundary) and thereby defines L. the algebraic results (write map, three-body mapping, Grassmannian structure) depend on pairwise overlap, not the tiling method. the geometric results (L, combinatorial ceiling, conservation on spatial cycles) depend on the Voronoi realization.

L is not a variational objective. the writing map drives the foam; L describes the resulting geometry. the active regime departs from minimality because perpendicular writes deposit structure in different directions. the resting state (no writes) is minimal because dL = 0.

L is bounded. U(d) is compact.

the combinatorial ceiling is exact. N unitaries cannot all be pairwise maximally distant. the achievable maximum is sqrt(N / (2(N-1))) of the theoretical maximum, depending only on N. derivation: Cauchy-Schwarz applied to norm(sum U_i)^2 >= 0.

L plausibly converges to 1/sqrt(2) of the theoretical maximum, under two hypotheses. the writing dynamics need to satisfy: (a) controllability — the write directions from overlapping observers collectively span the full Lie algebra, and (b) successive inputs are sufficiently decorrelated (channel_capacity.md: the mediation chain provides decorrelation along the chain).

both hypotheses are foam-geometry-dependent, not substrate facts. (a) depends on which observers exist and how their slices overlap; each observer’s writes live in only a 3D subspace of the d²-dimensional Lie algebra, so the spanning claim is collective, not per-observer. (b) cites the mediation chain, which provides decorrelation along the chain — whether that suffices for time-decorrelation at a single observer is the open question the README’s “mixing rate of the mediation chain” entry names.

under (a) and (b), on a compact group, a random walk whose step distribution generates the algebra converges to Haar measure. the expected pairwise distance under Haar is E[norm(W - I)_F] -> sqrt(2d) as d -> infinity (from E[norm(W - I)^2] = 2d, exact, and concentration of measure).

the Haar cost — the ratio E_Haar[L] / L_achievable — is sqrt((N-1)/N), exact and depending only on N.

the product: sqrt(N / (2(N-1))) * sqrt((N-1) / N) = 1/sqrt(2), independent of both N and d.

the two factors — the packing constraint and the saturation gap — are two halves of the same fraction.

the trace is the readout. trace_unique_of_kills_commutators: the trace is the unique scalar projection of a write. the overlap change tr(P * [W, P]) is visible on this channel. the observer has exactly one scalar readout, and it’s this one.

status

proven:

derived:

conditional (on controllability + decorrelation hypotheses, both foam-geometry-dependent):

cited:

observed:

bugs: