Extension from subsets closed under surviving edges
Let \(B\) be a sheaf over a commutative ring on an ordered graph with a finite vertex set. Write \(s(e),t(e)\) for the endpoints of an edge. Suppose \(B\) is flabby. Choose a set \(E_0\) of surviving edges, and assume that every edge module outside \(E_0\) is zero. If a vertex subset \(\Omega\) satisfies \[ e\in E_0,\ s(e)\in\Omega\Longrightarrow t(e)\in\Omega, \] every compatible section on \(\Omega\) extends to a global section.
Relation to the OpenAI notes
Relative extension through an explicit graph map.
OpenAI notes: The coset intersection is a finite relative lower ideal. A relative upper subset is closed under the surviving plane-labelled edges. Descending ambient extension gives relative flabbiness after localization. loc:relative-flabbiness.
Lean (primary): The same descending extension is factored into extend_closedUnder and a graph-pullback theorem. The coset map supplies vertex injectivity and explicit outgoing lifts; the root/unit criterion makes all outside edge modules zero. A separate theorem records selected-kernel equality under that pullback. Sheaves:106, MinimalExtensions:1520, PlaneLocalization:497, PlaneLocalization:591, TerminalKernels:393.
Role in the proof: Relative order determines the upper subsets on the coset, while the ambient graph supplies the extension. The localized stalks and edge maps remain the actual tensor-extended ambient modules, so selected kernels and their edge images retain their embeddings.
Proof
Step 1. Formulate the invariant maintained during extension
Let \(L\) be the set of surviving edges, and assume every edge outside \(L\) has zero edge module. We are given a compatible section on \(\Omega\), where \(s(e)\in\Omega,\ e\in L\) implies \(t(e)\in\Omega\).
We extend it one vertex at a time, retaining both this edge-closure property and all already assigned values. Induct on the finite number of vertices outside \(\Omega\).
Step 2. Choose the next vertex and use ordinary flabbiness
If no vertex remains, the assignment is global. Otherwise choose a maximal vertex \(a\) outside \(\Omega\). Every \(y>a\) lies in \(\Omega\), so the prescribed assignment restricts to a compatible section on the upper set \(\{y:y>a\}\).
Ordinary flabbiness supplies a global section \(h\) agreeing with those values. Define an ambient assignment \(g\) by \(g_a=h_a\) and \(g_y=f_y\) for \(y\ne a\), where \(f\) is the original assignment. Thus every value on \(\Omega\) is retained.
Step 3. Check every possible new edge condition
An edge whose two endpoints already lie in \(\Omega\) remains compatible.
For an edge \(a\to y\), maximality gives \(y\in\Omega\), and \(h_y\) equals the prescribed value there. Global compatibility of \(h\) proves the new equation.
An edge \(x\to a\) with \(x\in\Omega\) cannot survive: if it belonged to \(L\), edge closure of \(\Omega\) would force \(a\in\Omega\). Its edge module is therefore zero, so both restriction values equal zero automatically.
These exhaust all edges whose endpoints lie in \(\Omega\cup\{a\}\).
Step 4. Check closure and continue induction
For a surviving edge with source in the old \(\Omega\), its target was already in \(\Omega\). For one with source \(a\), its target is above \(a\), hence belongs to \(\Omega\) by maximality. Thus \(\Omega\cup\{a\}\) remains closed under surviving edges.
Exactly one missing vertex has been added, and all prescribed values were retained. Apply the induction hypothesis to this enlarged set and its compatible assignment. The resulting global section agrees with the new assignment on the enlarged set, and hence with the original assignment on \(\Omega\).
References
4. Localization at a root plane.
Lean formalization
OAI.KLInvariance.MomentGraph.Sheaf.extend_closedUnder: Sheaves.lean:106.