Lifting a partial morphism through the boundary image
Let \(k\) be a field and \(A\) a connected nonnegatively graded Noetherian domain over \(k\). Let \(B,C\) be graded sheaves on an ordered graph with finitely many edges. Assume both underlying sheaves are flabby and globally generated.
Let \(j\) be a degree-zero partial morphism on a vertex subset \(U\). Fix a vertex \(x\) with \[ \{y:x<y\}\subseteq U, \] and assume \(B^x\) is free. Write \[ r_B:B^x\to\partial B_x,\qquad r_C:C^x\to\partial C_x \] for the combined outgoing stalk maps, where \(\partial B_x=\prod_{e:x\to y}B^e\) and \(\partial C_x=\prod_{e:x\to y}C^e\). The known edge maps define \[ j_{\partial x}:\partial B_x\to\partial C_x,\qquad (v_e)_e\longmapsto(j_e(v_e))_e. \]
There exists a degree-zero map \(l_x:B^x\to C^x\) satisfying \[ r_C l_x=j_{\partial x}r_B. \]
Relation to the OpenAI notes
Graded extension preceding local decomposition.
OpenAI notes: The local-ring decomposition descends through the coset vertices, lifting stalk maps over the common boundary image and retaining both endpoint restrictions. loc:decomposition.
Lean (primary): The graded morphism-extension theorem works over a connected nonnegatively graded Noetherian domain. A homogeneous free-basis lift constructs the next stalk map, upper quotients construct the incoming edge maps, and finite-complement induction extends the partial morphism. SheafMorphisms:675, SheafMorphisms:723, SheafMorphisms:834.
Role in the proof: This graded extension is used in boundary recognition and splitting. The ungraded local-ring decomposition is a later construction with its own local-ring and model hypotheses.
Proof
Step 1. Map compatible assignments strictly above the new vertex
Every edge whose source is strictly above \(x\) also has target strictly above \(x\). Both vertices lie in \(U\).
For a section \(f\) above \(x\), define an ambient assignment \(g\) by \[ g_y= \begin{cases} j_y(f_y),&x<y,\\ 0,&x\not<y. \end{cases} \] On an edge \(e:y\to z\) with \(x<y\), the two partial-morphism squares give \[ \rho^{C,-}_e(j_yf_y) =j_e(\rho^{B,-}_ef_y) =j_e(\rho^{B,+}_ef_z) =\rho^{C,+}_e(j_zf_z). \] Thus \(g\) is a section of \(C\) strictly above \(x\).
Step 2. Map boundary tuples componentwise
Every outgoing target of \(x\) lies in \(U\), so its edge map \(j_e\) is part of the existing partial morphism. Applying these maps componentwise defines \(j_{\partial x}\).
Each \(j_e\) has degree zero, so \(j_{\partial x}\) preserves the product grading. Let \(b_B\) and \(b_C\) be the boundary maps from sections strictly above \(x\): their \(e\)-coordinates are the upper restrictions of the value at \(t(e)\). For the section \(f\) from Step 1 and its image \(g\), the upper commuting squares give \[ j_{\partial x}(b_B(f))=b_C(g). \] Hence \(j_{\partial x}\) sends the boundary image of \(B\) into that of \(C\).
Step 3. Replace the boundary images by the stalk-map images
The boundary-image theorem applies to each sheaf because both are flabby and globally generated. It gives \[ \operatorname{im}r_B=\operatorname{im}b_B, \qquad \operatorname{im}r_C=\operatorname{im}b_C. \] Together with Step 2, this yields \[ j_{\partial x}(\operatorname{im}r_B)\subseteq\operatorname{im}r_C. \]
Step 4. Restrict the target to a graded image
Let \(T=\operatorname{im}r_C\), with its induced grading. The inclusion from Step 3 defines a degree-zero map \[ g:B^x\to T,\qquad g(v)=j_{\partial x}(r_Bv). \] Also let \[ \pi:C^x\twoheadrightarrow T \] be \(r_C\) with its codomain restricted to its image. Every vector of \(T\) has an \(r_C\)-preimage, so \(\pi\) is surjective.
Step 5. Lift using a homogeneous free basis
The connected nonnegative Noetherian-domain assumptions supply a finite homogeneous basis \((m_i)\) of the free stalk \(B^x\). If \(m_i\) has degree \(d_i\), choose a preimage of \(g(m_i)\) under \(\pi\), then take its degree-\(d_i\) component. Since \(\pi\) preserves degree, the resulting vector \(v_i\) still satisfies \[ \pi(v_i)=g(m_i). \] Extend \(m_i\mapsto v_i\) linearly. The homogeneous lifting theorem proves that the resulting \(l_x\) has degree zero and \(\pi l_x=g\).
Finally, include \(T\) into \(\partial C_x\). This changes the last equation into \(r_C l_x=j_{\partial x}r_B\), as required.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.mapAssignment: SheafMorphisms.lean:613.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.compatible: SheafMorphisms.lean:625.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.boundary: SheafMorphisms.lean:639.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.boundaryMap: SheafMorphisms.lean:648.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.map_stalk_image_le: SheafMorphisms.lean:658.
OAI.KLInvariance.MomentGraph.GradedSheaf.OnHom.extend_stalk: SheafMorphisms.lean:675.