Graph maps, graph equivalences, and sheaf pullback
For ordered graphs \(G\) and \(H\), a graph map consists of maps on vertices and edges, \[ f_V:V_G\to V_H,\qquad f_E:E_G\to E_H, \] satisfying the two endpoint equations \[ f_V(s_G(e))=s_H(f_E(e)),\qquad f_V(t_G(e))=t_H(f_E(e)). \] A graph equivalence supplies bijections in place of these maps, with the same equations.
These data compare directed incidence. For example, on the chain \(0<1<2<3\) with only the edge \(0\to2\), exchanging the isolated vertices \(1,3\) and fixing the edge preserves incidence. The comparison \(1<2\) is sent to \(3>2\).
Pulling back a sheaf
For a sheaf \(B\) on \(H\), its pullback has \[ (f^*B)^x=B^{f_V(x)},\qquad (f^*B)^e=B^{f_E(e)}. \] The restrictions are those of \(B\), with their endpoint modules identified by the two graph-map equations. If \(B\) is graded, these same modules and maps retain their gradings.
An assignment pulls back by \[ (f^*m)_x=m_{f_V(x)}. \] On an edge \(e\), its compatibility equation is the original equation on \(f_E(e)\). Hence a global section pulls back to a global section. A prescribed stalk value can likewise be obtained by pulling back a section that generates the corresponding target stalk.
The label system pulls back as \(\alpha_G(e)=\alpha_H(f_E(e))\).
Reindexing by a graph equivalence
If both graph maps are bijective, the inverse maps give an inverse reindexing of assignments. They identify compatible assignments on a vertex subset with compatible assignments on its image.
Free stalks and upper quotients transport componentwise. To transport edge flabbiness, take a subset closed under all source edges. Its vertex image is closed under all target edges, because every target edge has a source-edge preimage and the endpoint equations agree. Extend the corresponding assignment there, then pull the extension back.
The finite boundary-existence construction uses this incidence transport when it identifies a formally adjoined bottom vertex with the removed vertex of the original graph.
Pullback with complete outgoing lifts
For a general graph map, the boundary-pullback theorem uses two additional conditions: \(f_V\) is injective, and every edge of \(H\) starting at \(f_V(x)\) is the image of an edge of \(G\) starting at \(x\).
These conditions make images of edge-closed subsets edge-closed and retain all outgoing kernel conditions. They are satisfied by the inclusion of an upper Bruhat interval.
References
The local source declarations below give the constructions and proofs.
Lean formalization
For both graph structures, vertex and edge are the maps, while source and target are the two incidence-preservation equations.
OAI.KLInvariance.MomentGraph.GraphMap: MinimalExtensions.lean:1457.
OAI.KLInvariance.MomentGraph.GraphEquiv: GradedSheaves.lean:270.
OAI.KLInvariance.MomentGraph.Sheaf.pullback: MinimalExtensions.lean:1473.
OAI.KLInvariance.MomentGraph.GradedSheaf.pullback: MinimalExtensions.lean:1650.
OAI.KLInvariance.MomentGraph.GradedSheaf.reindex: GradedSheaves.lean:279.
OAI.KLInvariance.MomentGraph.GradedSheaf.reindex_edgeFlabby: GradedSheaves.lean:336.
Used by
- The finite relative coset graph inside a Bruhat interval
- Localized labels and the coset sheaf
- Structure models under scalar extension and graph pullback
- Existence of normalized boundary sheaves on finite ordered graphs
- Construction of normalized models at every vertex
- Boundary-sheaf pullback with complete outgoing edge lifts
- Sheaf properties under pullback along surviving edges
- Selected kernels under outgoing-complete graph pullback