Sheaf isomorphisms and their partial forms
An isomorphism \(j:B\to C\) of sheaves on the same graph consists of linear isomorphisms on all stalks and edge modules such that \[ j_e\rho^{B,-}_e=\rho^{C,-}_ej_{s(e)},\qquad j_e\rho^{B,+}_e=\rho^{C,+}_ej_{t(e)}. \] Applying inverses to these equalities gives the commuting equations for \(j^{-1}\).
The component maps identify full assignments. The equations identify compatible assignments on every subset, so they give isomorphisms of the corresponding ambient section modules. They also identify the images and kernels formed from the restriction maps.
Isomorphisms on a subset
An isomorphism on \(U\subseteq V\) supplies \(j_x\) for \(x\in U\) and \(j_e\) for \(t(e)\in U\). Its upper commuting equation holds whenever \(t(e)\in U\), and its lower equation holds when both endpoints belong to \(U\).
It therefore identifies compatible tuples indexed on \(U\). In the ambient-assignment convention, this is the identification of the coordinates belonging to \(U\).
For \(U=V\), the data give a full isomorphism. For \(U=\{y:x<y\}\), they give an isomorphism above \(x\). Edges starting at \(x\) already have their edge isomorphisms because their targets lie above \(x\). The stalk isomorphism at \(x\) is the next object constructed in a descending extension.
Graded isomorphisms
A graded isomorphism requires \[ j_x(B^x_n)\subseteq C^x_n,\qquad j_e(B^e_n)\subseteq C^e_n \] for every degree and every component on which the maps are defined.
The inverse maps preserve degree as well. For a homogeneous \(c\in C_n\), write \(j^{-1}c=\sum_m b_m\) in homogeneous components. Then \(c=\sum_mj(b_m)\), with \(j(b_m)\in C_m\). Uniqueness of the homogeneous decomposition gives \(j(b_m)=0\) for \(m\ne n\). Injectivity of \(j\) gives \(b_m=0\) in those degrees, so \(j^{-1}c=b_n\).
This argument applies separately at each vertex and edge. Hence the stored forward grading conditions provide degree-preserving isomorphisms in both directions.
A one-edge partial isomorphism
For \(x\to y\), an isomorphism on \(\{y\}\) consists of \(j_y,j_e\) and the upper square. To extend it to the whole graph, choose \(j_x\) satisfying \[ j_e\rho^{B,-}_e=\rho^{C,-}_ej_x. \] The partial edge map makes this a specified equation for \(j_x\), which is solved by the boundary-cover argument.
References
The local definitions and constructions are listed below.
Lean formalization
The fields vertex, edge, lower, and upper describe the local isomorphisms and squares. The inherited toIso and toOnIso fields hold the ungraded data; vertex_graded and edge_graded assert the forward degree conditions.
OAI.KLInvariance.MomentGraph.Sheaf.Iso: GradedSheaves.lean:521.
OAI.KLInvariance.MomentGraph.Sheaf.OnIso: GradedSheaves.lean:970.
OAI.KLInvariance.MomentGraph.Sheaf.AboveIso: GradedSheaves.lean:692.
OAI.KLInvariance.MomentGraph.GradedSheaf.Iso: MinimalExtensions.lean:210.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnIso: MinimalExtensions.lean:214.
Used by
- Boundary models supported on principal lower ideals
- Uniqueness of the minimal-boundary sheaf up to graded isomorphism
- Local boundary models over a local coefficient ring
- Decomposing a free flabby sheaf into shifted boundary models
- A nonzero split summand of a primitive sheaf is the whole sheaf
- Selected kernels and scaled images under sheaf isomorphisms
- Descending decomposition into local boundary models
- Adding the model block at one unprocessed vertex
- Splitting a stalk while retaining all outgoing restrictions
- Structure models under scalar extension and graph pullback
- Boundary and stalk images under an isomorphism above a vertex
- Boundary-sheaf isomorphisms are detected at the normalized stalk
- Outgoing kernels under sheaf isomorphisms, sums, and shifts
- Transport of perfect section pairings through coefficient and sheaf isomorphisms
- Extending a partial sheaf isomorphism across one vertex
- Decomposing a stalk over an identified upper boundary