Sheaf properties preserved by lower-ideal extension
Let \(B\) be a graded sheaf on a lower ideal \(K\) of an ordered graph, and let \(\widetilde B\) be its extension by zero.
Free stalks, global generation, flabbiness, and the upper-quotient rule pass from \(B\) to \(\widetilde B\). If \(B\) is kernel-minimal at a vertex \(x\in K\), then so is \(\widetilde B\) at \(x\); every vertex outside \(K\) is automatically kernel-minimal.
Proof
Step 1. Identify the stalks and edge equations
Inside \(K\), the extended stalks identify with the old stalks. Outside \(K\), they are zero and hence free of rank zero.
An edge ending in \(K\) has both endpoints there and retains the old restriction equation. Every edge ending outside \(K\) has zero edge module, so its equation is the equality of two zero values. This proves that extension and restriction of assignments preserve the relevant compatibility conditions.
Step 2. Prove global generation
For a prescribed value at \(x\in K\), choose a global section of \(B\) with that value and extend it by zero. Step 1 makes it global on the larger graph.
If \(x\notin K\), its stalk has only the zero vector, realized by the zero section. Thus every extended stalk is generated.
Step 3. Extend compatible data on an upper set
Let \(U\) be upper in the full vertex poset, and let \(f\) be compatible on \(U\). Restrict its assignment to \(K\). It is compatible on \(U\cap K\), which is upper in the induced poset on \(K\): if \(x\in U\cap K\), \(x\le y\), and \(y\in K\), then \(y\in U\).
Flabbiness of \(B\) extends those values to a global section on \(K\). Extend that section by zero. It agrees with \(f\) on \(U\cap K\), and on \(U\setminus K\) both assignments have the unique zero value. This proves flabbiness of \(\widetilde B\).
Step 4. Preserve upper quotients
If an edge ends in \(K\), its upper stalk, edge module, and upper restriction are unchanged, so the quotient rule is inherited.
If it ends outside \(K\), its upper stalk and edge module are both zero. The map \(0\to0\) is surjective, and its kernel is \(0=\alpha_e0\). Thus the quotient rule holds on every edge.
Step 5. Preserve kernel minimality
For \(x\in K\), outgoing edges with target in \(K\) give the original conditions on \(B^x\). Those with target outside \(K\) have zero edge modules and impose a zero equation. Therefore the full outgoing kernel of \(\widetilde B\) identifies with the original kernel at \(x\).
If that kernel is contained in \(J B^x\), the same containment holds in \(J\widetilde B^x\). Outside \(K\), both the stalk and its kernel are zero, so minimality holds there as well.
References
The local source declarations below give the constructions and proofs.
Lean formalization
OAI.KLInvariance.MomentGraph.GradedSheaf.projectLower_compatible: MinimalExtensions.lean:575.
OAI.KLInvariance.MomentGraph.GradedSheaf.liftLower_compatible: MinimalExtensions.lean:581.
OAI.KLInvariance.MomentGraph.GradedSheaf.lowerExtend_free: MinimalExtensions.lean:563.
OAI.KLInvariance.MomentGraph.GradedSheaf.lowerExtend_generated: MinimalExtensions.lean:588.
OAI.KLInvariance.MomentGraph.GradedSheaf.lowerExtend_flabby: MinimalExtensions.lean:598.
OAI.KLInvariance.MomentGraph.GradedSheaf.lowerExtend_quotient: MinimalExtensions.lean:607.
OAI.KLInvariance.MomentGraph.GradedSheaf.lowerExtend_minimal: MinimalExtensions.lean:627.