Orientation invariance of the constrained summand: circuit flips #
For a fixed boundary-relative transition system κ, the corrected
constrained summand throughSummand is invariant under changing the
orientation, provided every internal flag on which the two
orientations disagree has an internal pairing partner — that is,
the difference set is supported on fully internal edges (equivalently,
on closed circuits; flips of path components are excluded).
The proof #
The difference set orientDiff o o' of two orientations is closed
under the matching (from both match_flips) and — under the
pairing-internality hypothesis — under the edge pairing (from both
pairing_flips). The colouring reindexing flipColouring applies
the odd-partner involution ∂ edge-wise on the difference set; it is
an involution of the core odd colourings fixing all boundary flags,
so the odd boundary constraint is preserved. At each vertex the
in-flags under o' are the unflipped in-flags under o together
with the matches of the flipped ones; each flipped visit contributes
a reversed pair block (one adjacent-swap sign) and trades the sign
factor ∂-partner-sign of the outgoing colour for that of the
incoming one. The two (−1)s per visit cancel, and the leftover
ratio sign(in) · sign(out) telescopes over the whole vertex product
to ∏_{f ∈ diff} sign(φ f), which is 1 because the difference set
is a disjoint union of full edges and the colouring is
pairing-constant.
Why the hypothesis is necessary #
Unrestricted orientation invariance is false. Counterexample
(ℓ = 2): one vertex v with two pendant edges {f₁, b₁},
{f₂, b₂} to boundary labels i₁, i₂, the matching f₁ ↔ f₂, and
odd state colours st i₁ = 0, st i₂ = 1. The boundary constraint
pins the unique contributing colouring, and the two orientations of
the path give summands proportional to −h(μ, {0, 3}) and
−h(μ, {1, 2}) respectively — different for generic h. The
per-visit sign ratio sign(x)·sign(y) telescopes to 1 only around
closed circuits; on a path it leaves the pinned end-colour signs, and
the ∂-reindexing moreover violates the pinned boundary values.
Consequently any orientation-independence interface must restrict to
circuit-supported differences (or fix path orientations by
convention).
Bool helpers #
The difference set of two orientations #
The internal flags on which two orientations of the same relative transition system disagree.
Equations
- RS.EdgeSubset.orientDiff o o' = {f ∈ F.internalFlags | o.isOut f ≠ o'.isOut f}
Instances For
Membership in the difference set: an internal flag the two orientations direct oppositely.
The difference set consists of internal flags.
Off the difference set, internal flags are oriented identically.
The difference set is closed under the matching.
The complement of the difference set is closed under the matching on internal flags.
Under the pairing-internality hypothesis, the difference set is closed under the edge pairing.
The complement of the difference set is closed under the pairing.
Boundary flags are never in the difference set.
The colouring reindexing #
The ∂-flip of a core odd colouring on the edges of the
difference set: the crux bijection for orientation invariance.
Equations
- RS.EdgeSubset.flipColouring o o' hpair φ = ⟨fun (g : ↥F.coreFlags) => if ↑g ∈ RS.EdgeSubset.orientDiff o o' then RS.oddPartner ℓ (↑φ g) else ↑φ g, ⋯⟩
Instances For
On the difference set the colouring is ∂-flipped.
Off it the colouring is unchanged.
The flip negates the incoming sign on the difference set: the
flip colouring is the colour flip on orientDiff o o'.
The flip leaves the incoming sign off the difference set alone.
The flip is an involution, so it is a bijection of the colouring sum.
The reindexing preserves the odd boundary constraint.
Vertex-local in-sets #
Signs as finset products #
The global sign telescopes #
The pair-list reindexing #
Assembly #
The orientation-invariance theorems #
Invariance under circuit flips: for a fixed relative transition system, the corrected constrained summand is invariant under changing the orientation, provided every internal flag on which the orientations disagree lies on a fully internal edge. (Unrestricted invariance is false: flipping a boundary-to-boundary path changes the summand — see the module docstring.)
Necessity of the hypothesis: a path-flip counterexample #
One vertex with two pendant edges to boundary labels 0 < 1, the
matching joining the two internal flags, and odd state colours 0
and 1 (with ℓ = 2). The odd boundary constraint pins the unique
contributing colouring; the two orientations of the resulting
boundary-to-boundary path give summands −1 and 0 for the
functional supported on the colour set {0, 3}.