Theorem Fixed stalks and the top normalization of adjoint projectors
✓ theorem proved

Fixed stalks and the top normalization of adjoint projectors

Fix a reduced word in a finite-generator Coxeter system and a basis of its real extended realization. Use its graded symmetric coefficient algebra and graded word sheaf \(W\), with the symmetric perfect word pairing. Let \(h:W\to W\) be a graded sheaf endomorphism, and let \(h^*\) be its descended adjoint.

At any vertex \(x\), \[ h_x=1\quad\Longleftrightarrow\quad h_x^*=1. \] In particular, if a normalized boundary sheaf \(B\) is split into \(W\) by \(i:B\to W,\ p:W\to B,\ pi=1\), then the projector \(e=ip\) and its adjoint satisfy \[ e_b=1_{W^b},\qquad e_b^*=1_{W^b} \] at the top vertex \(b\).

Proof

Step 1. Fix every word element supported at x

Let \(F\) be the word-module endomorphism induced by \(h\), and assume \(h_x=1\). For \(s\) supported at weight \(x\), the coordinate identity gives \[ \pi_x(Fs)=h_x(\pi_x(s))=\pi_x(s). \] At other weights both \(s\) and \(Fs\) have zero coordinates, because \(F\) preserves support. Their full evaluation tuples agree, and injectivity of evaluation gives \(Fs=s\).

Step 2. Test the difference made by the adjoint

For arbitrary \(m\) and any \(s\) supported at \(x\), symmetry and the adjoint equation give \[ \langle s,F^*m-m\rangle =\langle Fs,m\rangle-\langle s,m\rangle=0. \] The projection-kernel criterion therefore gives \[ \pi_x(F^*m-m)=0,\qquad \pi_x(F^*m)=\pi_x(m). \] By the descent formula, this is \(h_x^*(\pi_x(m))=\pi_x(m)\). The stalk projection is surjective, so \(h_x^*\) is the identity on the entire stalk.

Applying the same implication to \(h^*\), and using the double-adjoint identity, proves the converse.

Step 3. Reverse a split pair on rank-one modules

The boundary top stalk and the reduced-word top stalk are both graded isomorphic to \(A\). Use these identifications to view \(i_b,p_b\) as \(A\)-linear maps \(A\to A\).

Any such map is multiplication by its value at \(1\). Write these values as \(a,c\), so \(i_b(r)=ar\) and \(p_b(r)=cr\). The splitting \(p_bi_b=1\) gives \(ca=1\). Commutativity gives \(ac=1\), hence \(i_bp_b=1\).

Transporting back to the top stalk gives \(e_b=1_{W^b}\).

Step 4. Normalize the adjoint projector

Apply Steps 1–2 to the projector \(e\) at \(b\). Since \(e_b=1\), its adjoint also satisfies \(e_b^*=1\). Therefore the image of \(e^*\) has the whole word stalk at the top, with the same rank-one degree-zero normalization.

References

The source declarations below give the constructions and proofs.

Lean formalization

OAI.KLInvariance.TitsSpace.bottEndOfSheaf_fixed_support: BalancedPairings.lean:1466.

OAI.KLInvariance.TitsSpace.bottAdjointSheaf_fixed_stalk: BalancedPairings.lean:1478.

OAI.KLInvariance.TitsSpace.bottAdjointSheaf_fixed_stalk_iff: BalancedPairings.lean:1495.

OAI.KLInvariance.Graded.ModuleData.unit_split_reverse: Support.lean:39.

OAI.KLInvariance.TitsSpace.bottBoundarySplit_top_fixed: Support.lean:85.

OAI.KLInvariance.TitsSpace.bottBoundarySplitAdjoint_top_fixed: Support.lean:94.

Used by