Theorem R-reciprocity for the actual boundary-sheaf character
✓ theorem proved

R-reciprocity for the actual boundary-sheaf character

Fix a finite-generator Coxeter system and the polynomial coefficient algebra from a finite real basis of its extended realization. Let \(C_{x,b}(q)\) be the actual normalized boundary-stalk character on \(I_b=[1,b]\), and put \(d(x,b)=\ell(b)-\ell(x)\).

For every \(x\in I_b\), \[ \operatorname{reflect}_{d(x,b)}C_{x,b}(q) =\sum_{z\in I_b}R_{x,z}(q)C_{z,b}(q). \] The degree bound \(\operatorname{natDegree}C_{x,b}\le d(x,b)\) gives the equivalent Laurent expression \[ q^{d(x,b)}C_{x,b}(q^{-1}) =\sum_{x\le z\le b}R_{x,z}(q)C_{z,b}(q). \]

Relation to the OpenAI notes

Different intermediate statement.

OpenAI notes: The grading-normalization proposition identifies the stalk graded rank with the canonical KL polynomial and uses the Elias–Williamson character theorem through the cited moment-graph bridges. bg:normalization.

Lean (primary): The development defines the equality of actual stalk characters and canonical KL polynomials as StalkCharacterObligation. The unconditional theorem used in the main route is R-reciprocity for the actual character. The local strict half-degree implication retains an explicit nonzero-diagonal hypothesis. SheafMorphisms:219, DyerPaths:62, PositiveTransfer:186, CharacterReciprocity:1807, TerminalKernels:1273, TerminalKernels:1284.

Role in the proof: The comparison uses the actual character C and its proved R-reciprocity. Canonical P is handled by the later polynomial-rigidity induction. The local degree result applies to a diagonal coefficient satisfying its stated nonvanishing hypothesis.

Proof

Step 1. State the strong induction and its base

Induct strongly on \(\ell(b)\), proving the polynomial-reflection identity for every top of that length and every vertex below it.

If \(\ell(b)=0\), then \(b=1\) and the interval has one vertex. Its normalized character and diagonal R-polynomial are both \(1\), while \(\operatorname{reflect}_0(1)=1\). Thus the identity holds.

For the current \(b\), assume the result for every top \(c\) of strictly smaller length, at all vertices of \([1,c]\).

Step 2. Write both decompositions with one multiplicity family

Choose a reduced word \(\mathbf t\) for \(b\). For \(c,z\in I_b\), let \(L_z^c\) be the lower-character function, zero when \(z\not\le c\). Put \[ D_x^c=\operatorname{reflect}_{\delta(x,c)}L_x^c, \qquad \delta(x,c)=\max\{\ell(c)-\ell(x),0\}. \]

The simultaneous word decomposition supplies polynomials \(\mu_c\) with \[ C_z^{\mathbf t}=\sum_{c\in I_b}\mu_cL_z^c,\qquad K_x^{\mathbf t}=\sum_{c\in I_b}\mu_cD_x^c. \] The second identity comes from transporting the full outgoing kernels through the same graded sheaf decomposition. Each \(\mu_c\) records the same shifts in both identities.

Step 3. Compute the top multiplicity

Evaluate the first decomposition at \(z=b\). The reduced-word top character is \(1\). Lower-character normalization gives \(L_b^b=1\), and \(L_b^c=0\) for \(c\ne b\). Therefore \[ 1=C_b^{\mathbf t}=\mu_b. \]

Step 4. Substitute the word R-convolution

For the current reduced word, the word-character convolution gives \[ K_x^{\mathbf t}=\sum_{z\in I_b}R_{x,z}C_z^{\mathbf t}. \] Substituting both decompositions yields \[ \sum_c\mu_cD_x^c =\sum_zR_{x,z}\left(\sum_c\mu_cL_z^c\right). \]

Step 5. Verify each proper-center identity

Let \(c\ne b\). Since \(c\le b\), strict Bruhat comparison gives \(\ell(c)<\ell(b)\). The induction hypothesis supplies reciprocity on \([1,c]\).

The lower-character convolution theorem then gives, for the current \(x\in I_b\), \[ D_x^c=\sum_{z\in I_b}R_{x,z}L_z^c. \] Its proof includes both \(x\le c\) and \(x\not\le c\); in the latter case both sides are zero. Thus this identity is available for every proper center in the same ambient sum.

Step 6. Cancel the proper-center defects

All sums are finite. Interchange the two sums in Step 4 and bring them to one side: \[ 0=\sum_{c\in I_b}\mu_c \left(D_x^c-\sum_{z\in I_b}R_{x,z}L_z^c\right). \] Every summand with \(c\ne b\) vanishes by Step 5. The remaining coefficient is \(\mu_b=1\), so \[ D_x^b=\sum_zR_{x,z}L_z^b. \] For the top center, \(L_z^b=C_{z,b}\) and \(D_x^b=\operatorname{reflect}_{d(x,b)}C_{x,b}\). This is the required induction conclusion. It is the source's finite-character cancellation with distinguished index \(b\).

Step 7. Interpret the reciprocal formula

Boundary kernel duality gives \(\operatorname{natDegree}C_{x,b}\le d(x,b)\). The bounded-reflection formula therefore identifies the left side with \(q^{d(x,b)}C_{x,b}(q^{-1})\).

The R-polynomial vanishes unless \(x\le z\), so only \(x\le z\le b\) contribute. On the one-edge interval \([1,s]\), the lower-vertex identity is \[ q=R_{1,1}+R_{1,s}=1+(q-1), \] matching the kernel character of the degree-one root ideal.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.BruhatGraph.stalkCharacter_reciprocity: DyerPaths.lean:62.

OAI.KLInvariance.BruhatGraph.stalkCharacter_natDegree_le: EdgeLocalization.lean:364.

Used by