Theorem Symmetric perfect pairings on the actual boundary sheaf
✓ theorem proved

Symmetric perfect pairings on the actual boundary sheaf

Fix a Coxeter system with finitely many simple generators, an element \(b\), and a basis of the real extended realization \(V\). Let \[ A=\mathbb R[X_1,\ldots,X_N] \] be the resulting polynomial coefficient algebra, with variables in degree one, and let \(B_A\) be the chosen normalized boundary sheaf on \([1,b]\).

Its global section module has an \(A\)-bilinear symmetric perfect pairing with support orthogonality and degree shift \(\ell(b)\): \[ \langle\Gamma(B_A)_i,\Gamma(B_A)_j\rangle \subseteq A_{i+j-\ell(b)}. \]

Relation to the OpenAI notes

Different construction of the boundary duality.

OpenAI notes: The normalization proof uses the normalized self-dual projective and the cited moment-graph/Soergel equivalence to relate the stalk and lower-support layer, before translating grading conventions. bg:normalization.

Lean (primary): The word pairing is constructed by tensor induction. Supported weight submodules prove adjoint kernel preservation, which descends the adjoint to a graded sheaf map. The adjoint image is dual to the original boundary summand; primitive recognition and the canonical comparison isomorphism produce its perfect symmetric pairing. BalancedPairings:717, EdgeLocalization:833, EdgeLocalization:906, EdgeLocalization:1006.

Role in the proof: The proof follows the integral adjoint, image-section duality, primitive recognition, and canonical-comparison maps. The trace lowers degree once per word letter. Coefficient transport and a boundary-recognition isomorphism deliver the pairing on the chosen polynomial-coordinate sheaf.

Proof

Step 1. Construct the symmetric-coefficient boundary summand

Write \(R=\operatorname{Sym}_{\mathbb R}(V)\), with the same degree-one grading, and choose a reduced word \(\mathbf s\) for \(b\). Its length is \(\ell(b)\).

Let \(B_R\) be the normalized boundary sheaf over \(R\), and let \(W\) be the graded word sheaf of \(\mathbf s\). The word theorems supply free stalks, the upper-quotient rule, flabbiness, and global generation for \(W\). Its top stalk is \(R\) in degree zero.

The normalized top identifications and the top-splitting theorem give \[ i:B_R\to W,\qquad p:W\to B_R,\qquad pi=1. \] The word module \(M=M_{\mathbf s}\) is isomorphic to \(\Gamma(W)\), by the word lifting theorem and injectivity of full evaluation. We use this identification for the induced maps on sections, writing \(i:\Gamma(B_R)\to M\) and \(p:M\to\Gamma(B_R)\).

The tensor construction gives a symmetric perfect pairing \(D\) on \(M\). The word-degree theorem gives its shift \(|\mathbf s|=\ell(b)\).

Step 2. Construct and recognize the adjoint image

The sheaf projector \(e=ip\) induces the module projector \(F=ip\) on \(M\). The integral adjoint \[ F^*=D^{-1}F^\vee D \] descends to a graded sheaf projector \(e^*\), by the supported-stalk annihilator and quotient-descent argument.

Let \(C=\operatorname{im}e^*\), with inclusion \(j:C\to W\) and projection \(q:W\to C\). The adjoint-image recognition theorem verifies its inherited free stalks, upper quotients, flabbiness, global generation, primitive identity, and normalized top, and gives \(C\simeq B_R\).

The canonical-comparison theorem proves that the specified map \[ h=qi:B_R\xrightarrow{\sim}C \] is an isomorphism. Its inclusion satisfies \(jh=e^*i\).

Step 3. Identify the adjoint-image sections with the dual

The image-section theorem identifies \[ \Gamma(C)\simeq\operatorname{im}F^*. \] For the split module maps \(i,p\), the adjoint-summand theorem gives a perfect pairing between \(\operatorname{im}F^*\) and \(\Gamma(B_R)\): \[ Q(c,n)=\langle j(c),i(n)\rangle_M. \] Here both included sections are viewed as word elements.

Its associated linear isomorphism is \(\Gamma(C)\to\Gamma(B_R)^*\). To represent a functional \(\lambda\), first form \(\lambda\circ p\) on \(M\), then take \(D^{-1}(\lambda\circ p)\). The adjoint-summand theorem places this element in \(\operatorname{im}F^*\) and proves uniqueness.

Step 4. Pull back through the canonical comparison

Define \[ P(m,n)=Q(h(m),n),\qquad m,n\in\Gamma(B_R). \] The map \(h\) is a section isomorphism and the map associated to \(Q\) is an isomorphism to the dual. Their composite proves perfection of \(P\).

The canonical formula identifies this pairing explicitly: \[ \begin{aligned} P(m,n) &=\langle F^*i(m),i(n)\rangle_M\\ &=\langle i(m),Fi(n)\rangle_M\\ &=\langle i(m),i(n)\rangle_M. \end{aligned} \] The last equality follows from \(Fi=ipi=i\). Symmetry of the word form therefore gives symmetry of \(P\).

Step 5. Verify the degree and support equations

The inclusion \(i\) preserves degree. For inputs of degrees \(r,s\), the displayed formula and the word-degree theorem give \[ P(m,n)\in R_{r+s-\ell(b)}. \]

If \(m\) is supported at a vertex \(x\), the included section has zero value at every other vertex because each inclusion component is linear. If \(n_x=0\), its included section also has zero \(x\)-value. Word support orthogonality gives \(P(m,n)=0\). Symmetry gives the right support equation as well.

Thus \(P\) is a symmetric perfect support-orthogonal section pairing over \(R\).

Step 6. Transfer the result to the chosen polynomial sheaf

The chosen basis gives a graded algebra isomorphism \[ \varphi:R\xrightarrow{\sim}A. \] It sends every symmetric root label to the corresponding polynomial label. The coefficient-transport recognition theorem gives a graded sheaf isomorphism \[ \psi:B_A\xrightarrow{\sim}{}^\varphi B_R. \]

Transport \(P\) along \(\varphi\), then pull it back along \(\psi\). Explicitly, if \(\overline\psi(m)\) denotes the underlying \(R\)-section corresponding to \(\psi(m)\), set \[ \langle m,n\rangle_A =\varphi\!\left(P(\overline\psi(m),\overline\psi(n))\right). \] The pairing-transport theorem proves perfection, symmetry, both support equations, and preservation of the shift. This produces the required pairing on the actual chosen polynomial-coordinate boundary sheaf \(B_A\).

References

The source declarations below give the constructions and proofs.

Lean formalization

OAI.KLInvariance.BruhatGraph.boundarySheaf_symmetricPairing: EdgeLocalization.lean:1006.

OAI.KLInvariance.TitsSpace.bottCanonicalAdjointIso: EdgeLocalization.lean:833.

OAI.KLInvariance.TitsSpace.bottCanonicalPerfectPairing_apply: EdgeLocalization.lean:906.

OAI.KLInvariance.TitsSpace.symmetricBoundarySheaf_symmetricPairing_word: EdgeLocalization.lean:977.

OAI.KLInvariance.TitsSpace.symmetricBoundarySheaf_symmetricPairing: EdgeLocalization.lean:988.

Used by