Theorem Supported sections are dual to a generated stalk
✓ theorem proved

Supported sections are dual to a generated stalk

Let \(R\) be a commutative domain, \(A_y\) torsion-free \(R\)-modules, and \(M\subseteq\prod_y A_y\). Let \(p\ne0\) define integral cuts \(C_y m=p\delta_y(m_y)\in M\). Suppose \(D:M\xrightarrow{\sim}M^*\) gives a perfect pairing balanced for these cuts.

Fix \(x\) and assume that \(\pi_x:M\to A_x\), \(m\mapsto m_x\), is surjective. Then there is an \(R\)-linear isomorphism \[ \Phi_x:M_{\{x\}}\xrightarrow{\sim}A_x^*. \] It is characterized by \[ \Phi_x(m)(n_x)=\langle m,n\rangle\qquad(m\in M_{\{x\}},\ n\in M). \] Its inverse is \[ \Psi_x(\lambda)=D^{-1}(\lambda\circ\pi_x). \]

For a globally generated sheaf with torsion-free stalks, nonzero annihilating labels, finitely many edges, and a balanced perfect section pairing, these hypotheses hold with \(M=\Gamma(B)\) and \(A_x=B^x\).

Relation to the OpenAI notes

Different construction of the stalk–costalk relation.

OpenAI notes: The normalization proof describes a lower-support layer, relates it to the outgoing kernel using flabbiness, and obtains its graded rank through the cited moment-graph and Soergel-module results. It temporarily doubles degrees and uses the shift convention M(k)^j = M^{j+k}. bg:normalization, eq:bg-upward-rank.

Lean (primary): The local construction starts with a free sheaf and a specified balanced perfect section pairing. Integral cuts identify vertex support with an evaluation-kernel annihilator. The incoming label product identifies that support module with the outgoing kernel, giving an explicit linear equivalence with the stalk dual. Support:521, CharacterReciprocity:150, CharacterReciprocity:185.

Role in the proof: The Blueprint follows these integral maps. The pairing construction and its homogeneity equation supply the later graded reciprocity statement. Degree shifts use the Lean convention M(d)n = M{n-d}.

Proof

Step 1. Construct the supported representative of a stalk functional

For \(\lambda\in A_x^*\), the composite \(\lambda\circ\pi_x\) is an \(R\)-linear functional on \(M\). Perfection supplies a unique \(m=D^{-1}(\lambda\circ\pi_x)\).

If \(n_x=0\), then \[ \langle m,n\rangle=(\lambda\circ\pi_x)(n)=\lambda(0)=0. \] The support-annihilator theorem applies to our domain, torsion-free coordinate modules, nonzero cut scalar, and balanced perfect pairing. It implies \(m\in M_{\{x\}}\). Thus \(\Psi_x\) has the asserted codomain. It is linear because precomposition by \(\pi_x\) and \(D^{-1}\) are linear.

Step 2. Descend a supported section to a stalk functional

Let \(m\in M_{\{x\}}\). For \(a\in A_x\), choose \(n\in M\) with \(n_x=a\), using surjectivity, and set \[ \Phi_x(m)(a)=\langle m,n\rangle. \] If \(n'\) is another lift, \((n-n')_x=0\). The support-annihilator theorem gives \[ \langle m,n\rangle-\langle m,n'\rangle =\langle m,n-n'\rangle=0. \] Hence the value is independent of the lift.

If \(n,n'\) lift \(a,a'\), then \(n+n'\) lifts \(a+a'\), so bilinearity proves additivity in \(a\). Likewise \(rn\) lifts \(ra\), giving scalar linearity. Using the same lift for two supported first arguments proves that \(\Phi_x\) itself is \(R\)-linear.

Step 3. Check both compositions

For \(\lambda\in A_x^*\) and \(a\in A_x\), choose a lift \(n\). Then \[ \Phi_x(\Psi_x(\lambda))(a) =D(D^{-1}(\lambda\circ\pi_x))(n) =\lambda(n_x)=\lambda(a). \] Thus \(\Phi_x\Psi_x\) is the identity.

For \(m\in M_{\{x\}}\), the defining formula gives \(\Phi_x(m)\circ\pi_x=D(m)\) as functionals on \(M\). Applying \(D^{-1}\) yields \[ \Psi_x(\Phi_x(m))=D^{-1}(D(m))=m. \] Both inverse identities hold, proving the isomorphism.

Step 4. Specialize to graph sheaves

For a sheaf, choose \(p=\prod_e\alpha_e\). Each label annihilates its edge module, so the integral-cut construction makes \(C_y\) an endomorphism of \(\Gamma(B)\). The product is nonzero because the coefficient ring is a domain and the finite collection of labels is nonzero. Global generation is precisely the surjectivity of every evaluation map. Substitution into Steps 1–3 gives \[ \Gamma(B)_{\{x\}}\simeq\operatorname{Hom}_R(B^x,R). \]

References

The source declarations below give the definitions and constructions used here.

Lean formalization

OAI.KLInvariance.IntegralSupport.dualToSupport: Support.lean:541.

OAI.KLInvariance.IntegralSupport.dualToSupport_bijective: Support.lean:551.

OAI.KLInvariance.IntegralSupport.supportEquivDual: Support.lean:570.

OAI.KLInvariance.MomentGraph.Sheaf.singleSupportEquivDual: Support.lean:856.

Used by