Splitting a stalk while retaining all outgoing restrictions
Let \(R\) be a commutative local ring with maximal ideal \(\mathfrak m\), and let \(B,C\) be sheaves on an ordered graph. Fix an isomorphism \(j\) above a vertex \(x\). Write \(d_x^B,d_x^C\) for the full outgoing stalk maps and \(j_e\) for its edge isomorphisms.
Assume both sheaves are globally generated and flabby, both stalks \(B^x,C^x\) are finite free, and \[ \ker d_x^C\subseteq\mathfrak mC^x. \] Then there is a linear map \(g:B^x\to C^x\) and an isomorphism \[ e_x:B^x\xrightarrow{\sim}C^x\oplus Q,\qquad Q=\ker g, \] where \(Q\) is finite free, such that for every outgoing edge \(e\) and \(v\in B^x\), \[ j_e(\rho^{B,-}_e(v)) =\rho^{C,-}_e((e_x(v))_1). \] Thus the first summand retains every prescribed outgoing restriction.
Relation to the OpenAI notes
The same single-lift local-ring argument.
OpenAI notes: The local decomposition lifts the given stalk map through the existing model stalk cover. Minimality makes the lift surjective after reduction modulo the maximal ideal; Nakayama gives surjectivity, the free target supplies a section, and the finite projective kernel is free over the local ring. loc:decomposition.
Lean (primary): LocalCover first proves the surjectivity of a single lift with a finite target and the displayed kernel condition. It then constructs the lift and right inverse, gives the explicit splitting map and proves the kernel finite projective and free. AboveIso.extend_stalk constructs the common outgoing image and applies this algebraic theorem. ModuleAlgebra:165, ModuleAlgebra:206, GradedSheaves:766.
Role in the proof: The Blueprint follows the single-lift proof and writes the cover triangle, splitting inverse, kernel projection and each outgoing restriction equality. The graded two-lift argument belongs to the separate graded construction.
Proof
Step 1. Identify the common outgoing image
Put \(\Delta_x(B)=\prod_{e:s(e)=x}B^e\), and define \(\Delta_x(C)\) in the same way. Let \(J:\Delta_x(B)\simeq\Delta_x(C)\) be the product edge isomorphism. The boundary-image transport theorem gives \[ J(\operatorname{im}d_x^B)=\operatorname{im}d_x^C. \] Put \(N=\operatorname{im}d_x^C\), viewed as an actual submodule of \(\Delta_x(C)\).
Define \[ \pi:C^x\to N,\qquad \pi(c)=d_x^C(c), \] with its codomain restricted to the image. This map is surjective by the definition of \(N\).
Step 2. Construct the second surjection onto that image
Define \[ \varphi:B^x\to N,\qquad \varphi(b)=J(d_x^B(b)). \] Step 1 shows that every value lies in \(N\), so the codomain restriction is defined.
For \(n\in N\), the same image equality supplies \(b\in B^x\) with \(J(d_x^B(b))=n\). Thus \(\varphi\) is onto.
The inclusion \(N\hookrightarrow\Delta_x(C)\) is injective. Therefore \[ \ker\pi=\ker d_x^C\subseteq\mathfrak mC^x. \]
Step 3. Split the two covers of the common image
The two source modules \(B^x,C^x\) are finite free, \(\pi,\varphi\) are onto, and \(\pi\) has the required kernel condition.
The local cover-splitting theorem gives a surjective lift \(g:B^x\to C^x\), a finite free kernel \(Q\), and an isomorphism \(e_x\) with \[ \pi((e_x(b))_1)=\varphi(b). \]
Step 4. Evaluate the equality at each edge
Both sides of the last equality are elements of the submodule \(N\subseteq\Delta_x(C)\). Include them in the boundary module and evaluate at an outgoing edge \(e\).
The left side becomes \(\rho^{C,-}_e((e_x(b))_1)\). The right side becomes \(j_e(\rho^{B,-}_e(b))\), by the coordinate formula for \(J\). This is the required commuting equality for every outgoing edge.
The kernel \(Q\) therefore supplies the new free directions while the \(C^x\)-component carries all outgoing data already determined above \(x\).
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.MomentGraph.Sheaf.AboveIso.extend_stalk: GradedSheaves.lean:766.