The full outgoing kernel is dual to the stalk
Let \(B\) be a free sheaf on a Bruhat interval with finitely many edges and its polynomial coefficient ring \(A\). Its stalks are finite free, its upper maps realize the label quotients, and its global sections generate every stalk.
Suppose \(B\) has a section duality \(D:\Gamma(B)\xrightarrow{\sim}\Gamma(B)^*\), balanced for the cuts made with the product of all edge labels. Write \[ K_x=\ker\!\left(B^x\longrightarrow\prod_{e:\,s(e)=x}B^e\right), \qquad \beta_x=\prod_{e:\,t(e)=x}\alpha_e. \] Then the full outgoing kernel, also called the costalk here, has an \(A\)-linear isomorphism \[ \kappa_x:K_x\xrightarrow{\sim}\operatorname{Hom}_A(B^x,A). \] In particular, \(K_x\) is finite free.
The isomorphism is determined by \[ \kappa_x(k)(n_x)=\langle\delta_x(\beta_x k),n\rangle \quad(k\in K_x,\ n\in\Gamma(B)). \]
With a homogeneous section pairing, the degrees of these maps produce the graded reciprocity formula developed in The grading shift in stalk–costalk reciprocity.
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. Apply the incoming-product isomorphism
The theorem Incoming labels identify the outgoing kernel with vertex support applies to the given free sheaf and its polynomial labels. It supplies \[ \gamma_x:K_x\xrightarrow{\sim}\Gamma(B)_{\{x\}}, \qquad k\longmapsto\delta_x(\beta_x k). \] Its proof checks all incoming and outgoing edge equations, divides a supported stalk value by the product of incoming primes, and cancels that product in each outgoing edge module.
Step 2. Apply supported-stalk duality
The coefficient ring \(A\) is a domain. Each free stalk is torsion-free, and every polynomial label is nonzero. The finite label product \(p\) is therefore nonzero. Upper quotients imply label annihilation on edge modules and hence define the cuts on global sections.
The supplied duality is perfect and balanced. Finally, global generation says that \(\pi_x:\Gamma(B)\to B^x\) is surjective. These are exactly the hypotheses of Supported sections are dual to a generated stalk. It gives \[ \Phi_x:\Gamma(B)_{\{x\}}\xrightarrow{\sim}(B^x)^*, \qquad \Phi_x(m)(n_x)=\langle m,n\rangle. \]
Step 3. Compose the maps and identify the inverse
Set \(\kappa_x=\Phi_x\gamma_x\). Both maps are linear isomorphisms, and substitution of \(\gamma_x(k)=\delta_x(\beta_x k)\) gives the stated formula.
For a functional \(\lambda\in(B^x)^*\), first form \[ m=D^{-1}(\lambda\circ\pi_x). \] Supported-stalk duality proves that \(m\) is supported at \(x\). The incoming-product theorem then gives a unique \(k\in K_x\) with \(m_x=\beta_x k\). This \(k\) is \(\kappa_x^{-1}(\lambda)\).
The first construction evaluates \(\lambda\) on every \(n_x\), so it recovers \(\lambda\). Conversely, starting with \(k\), perfection recovers \(\gamma_x(k)\), and injectivity of multiplication by \(\beta_x\) recovers \(k\). Thus the explicit formulas are mutually inverse.
Step 4. Transport a dual basis
Choose a finite free basis \(v_1,\ldots,v_r\) of \(B^x\). Its coordinate functionals \(v_i^*\) satisfy \(v_i^*(v_j)=1\) for \(i=j\) and \(0\) otherwise.
Any functional \(\lambda\) equals \(\sum_i\lambda(v_i)v_i^*\), because both sides agree on every basis vector. These coordinate functionals are linearly independent by evaluation on the \(v_j\). Hence they form a finite free basis of \((B^x)^*\).
Pulling this basis back along \(\kappa_x\) gives a finite free basis of \(K_x\).
Step 5. Compute the isomorphism on one edge
Use \(A=\mathbb R[\alpha]\) and the section pairing \[ \langle(f,g),(f',g')\rangle=\frac{ff'-gg'}{2\alpha}. \] At the lower vertex, \(K_x=\alpha A\) and \(\beta_x=1\). For \(k=\alpha h\), \[ \kappa_x(\alpha h)(f)=\langle(\alpha h,0),(f,g)\rangle=hf/2. \] Thus the inverse sends the stalk functional \(f\mapsto cf\) to \(2\alpha c\).
At the upper vertex, \(K_y=A\) and \(\beta_y=\alpha\). For \(k\in A\), \[ \kappa_y(k)(g)=\langle(0,\alpha k),(f,g)\rangle=-kg/2. \] The inverse sends \(g\mapsto cg\) to \(-2c\). This exhibits both costalk isomorphisms and the incoming factor at the upper vertex.
References
The source declarations below give the definitions and constructions used here.
Lean formalization
OAI.KLInvariance.BruhatGraph.FreeSheaf.costalkEquivDual: CharacterReciprocity.lean:185.
OAI.KLInvariance.BruhatGraph.FreeSheaf.costalk_free_of_duality: CharacterReciprocity.lean:197.
OAI.KLInvariance.BruhatGraph.costalkEquivDual: Support.lean:1447.
OAI.KLInvariance.BruhatGraph.costalk_free_of_duality: Support.lean:1459.