Sections supported at a vertex and their stalk values
Let \(R\) be a commutative ring, let \(V\) be a set, and let \(A_x\) be an \(R\)-module for every \(x\in V\). Suppose \[ M\subseteq\prod_{x\in V}A_x \] is an \(R\)-submodule of assignments. Coordinate evaluation is the linear map \[ \pi_x:M\to A_x,\qquad m\mapsto m_x. \] The support of an assignment is the set of its nonzero coordinates. The submodule supported at \(x\) is \[ M_{\{x\}}=\{m\in M:m_y=0\text{ for every }y\ne x\}. \] It contains the zero assignment. Addition and scalar multiplication preserve the displayed coordinate equations.
Write \(\delta_x(v)\) for the assignment with value \(v\) at \(x\) and zero elsewhere. For a graph sheaf \(B\), take \(M=\Gamma(B)\) and \(A_x=B^x\). The supported stalk is \[ S_x=\{v\in B^x:\delta_x(v)\in\Gamma(B)\}. \] Thus \(S_x\) is a submodule of a single stalk, whereas \(\Gamma(B)_{\{x\}}\) is a submodule of the global section module.
Evaluation and zero extension
The linear maps \[ S_x\longrightarrow\Gamma(B)_{\{x\}},\quad v\longmapsto\delta_x(v), \qquad \Gamma(B)_{\{x\}}\longrightarrow S_x,\quad m\longmapsto m_x \] are inverse. Evaluating \(\delta_x(v)\) gives \(v\). Conversely, \(m\) supported at \(x\) has the same value as \(\delta_x(m_x)\) at \(x\), and both vanish elsewhere.
The incident-edge equations
For \(\delta_x(v)\) to be a global section, its two restrictions must agree on every edge. There are three cases.
- On an outgoing edge \(x\to y\), the upper value is zero, so \(\rho^-_e(v)=0\).
- On an incoming edge \(z\to x\), the lower value is zero, so \(\rho^+_e(v)=0\).
- On an edge with neither endpoint \(x\), both values are zero.
If \(K_x\) is the full outgoing kernel, these conditions give \[ S_x=K_x\cap\bigcap_{e:\,t(e)=x}\ker\rho^+_e. \] With the upper-quotient condition, the incoming kernels are \(\alpha_e B^x\).
For the one-edge congruence sheaf over \(\mathbb R[\alpha]\), \[ K_x=\alpha A,\quad K_y=A,\qquad S_x=\alpha A,\quad S_y=\alpha A. \] At the lower vertex the outgoing equation imposes divisibility by \(\alpha\); at the upper vertex the incoming equation imposes it. The supported sections are respectively \((\alpha h,0)\) and \((0,\alpha h)\).
Functionals vanishing on a submodule
For \(N\subseteq M\), its dual annihilator is \[ N^\perp=\{\lambda\in\operatorname{Hom}_R(M,R): \lambda(n)=0\text{ for every }n\in N\}. \] In particular, \((\ker\pi_x)^\perp\) consists of functionals whose value is unchanged when one changes a section without changing its \(x\)-coordinate. The support-annihilator theorem identifies this module with \(M_{\{x\}}\) through a balanced perfect pairing.
References
The source declarations below give the definitions and constructions used here.
Lean formalization
OAI.KLInvariance.IntegralSupport.projection: Support.lean:477.
OAI.KLInvariance.IntegralSupport.supported: Support.lean:479.
OAI.KLInvariance.MomentGraph.Sheaf.singleStalk: Support.lean:996.
OAI.KLInvariance.MomentGraph.Sheaf.singleStalkEquivSupport: Support.lean:999.
OAI.KLInvariance.MomentGraph.Sheaf.single_section_iff: Support.lean:898.