Decomposing a stalk over an identified upper boundary
Work over a connected nonnegatively graded Noetherian domain. Let \(B,C\) be graded sheaves on an ordered graph with finitely many edges. Fix a vertex \(x\).
Assume both sheaves are flabby and globally generated, both stalks \(B^x,C^x\) are free, and \(C\) is kernel-minimal at \(x\). Suppose an isomorphism above \(x\) is given, with degree-preserving edge components. Let \[ T_x:\partial_xB\xrightarrow{\sim}\partial_xC \] be its induced boundary-product isomorphism.
Then there are a degree-zero map \(g:B^x\to C^x\), a finite graded free module \(K=\ker g\), and a graded isomorphism \[ E:B^x\xrightarrow{\sim}C^x\oplus K \] such that \[ T_xr_x^B(v)=r_x^C\bigl(\operatorname{pr}_1E(v)\bigr). \]
Proof
Step 1. Identify the target boundary image
The above-\(x\) isomorphism identifies compatible tuples on the strict upper set. Applying its edge maps to their quotient values identifies the boundary images: \[ T_x(B^{\delta x})=C^{\delta x}. \] Use \(T=\operatorname{im}r_x^C=C^{\delta x}\) as a common target. Its grading is the range grading of the degree-zero stalk map \(r_x^C\). The boundary product is finite because the edge set is finite. This supplies the graded target directly from the finite stalk and edge modules.
Step 2. Form the two free covers
Global generation and flabbiness give the boundary-image equality for both sheaves. Therefore \[ f=T_xr_x^B:B^x\twoheadrightarrow T,\qquad \pi=r_x^C:C^x\twoheadrightarrow T \] are surjections with codomains restricted to that image. They preserve degree because the stalk restrictions and the edge components of \(T_x\) preserve degree.
The source stalks are finite free. Kernel minimality of \(C\) says \(\ker\pi\subseteq JC^x\). Thus \(\pi\) is a minimal graded free cover of \(T\).
Step 3. Apply the free-cover remainder theorem
That theorem supplies degree-zero maps \(g:B^x\to C^x\) and \(h:C^x\to B^x\) with \[ \pi g=f,\qquad gh=1. \] For the finite graded free remainder \(K=\ker g\), it gives \[ E(v)=(gv,v-hgv). \] The inverse is \((c,k)\mapsto hc+k\), with both maps preserving degree.
Step 4. Check the boundary equation
The first component of \(E(v)\) is \(gv\). Hence \[ r_x^C(\operatorname{pr}_1E(v)) =\pi(gv)=f(v)=T_xr_x^B(v). \] The remainder contributes zero to this boundary value. It is the part represented by new model summands centered at \(x\) in the global decomposition.
References
The local source declarations below give the constructions and proofs.
Lean formalization
OAI.KLInvariance.MomentGraph.Sheaf.AboveIso.split_stalk_graded: SheafMorphisms.lean:1496.