Recognition of the adjoint boundary image
Fix a reduced word \(\mathbf s\) with product \(b\), and work on the interval \([1,b]\) over the symmetric coefficient algebra of a finite-generator real extended realization. Let \(W\) be the graded word sheaf and \(B\) a normalized boundary sheaf with the same labels.
There exists a graded splitting \(i:B\to W,\ p:W\to B,\ pi=1_B\). For any such splitting, put \(e=ip\), let \(e^*\) be the word-sheaf adjoint, and set \[ C=\operatorname{im}e^*. \] Then \(C\) is graded isomorphic to \(B\).
Proof
Step 1. Supply the splitting from the normalized top
The word-stalk theorem gives free stalks for \(W\), and its edge theorem gives the upper-quotient rule. Word flabbiness and the section construction give flabbiness and global generation.
The reduced-word top-stalk theorem identifies \(W^b\) with the degree-zero unit module \(A\). The boundary normalization identifies \(B^b\) with the same module. Composing these identifications gives inverse degree-zero maps \[ a_0:B^b\to W^b,\qquad b_0:W^b\to B^b. \] The interval is finite with greatest vertex \(b\), and the symmetric coefficient algebra is a connected graded Noetherian domain. The top-splitting theorem therefore supplies a splitting of \(B\) into \(W\).
The word-boundary summand theorem records this application with all its ring, graph and sheaf hypotheses.
Fix any splitting \(i,p\) for the remaining steps.
Step 2. Form the adjoint image and its actual split maps
The word adjoint is a graded sheaf endomorphism, and the adjoint of an idempotent is idempotent. Thus \(e^*\) is a graded sheaf projector.
Its image sheaf has the component images as stalks and edge modules. Let \[ j:C\to W,\qquad q:W\to C \] be its inclusion and projection. They satisfy \(qj=1_C\) and \(jq=e^*\).
Step 3. Transfer the sheaf properties to C
The split-summand theorem transfers the upper-quotient rule, flabbiness, and global generation from \(W\) to \(C\). Each stalk of \(C\) is a graded direct summand of a finite free stalk of \(W\), hence projective.
The coefficient algebra is connected, nonnegative, Noetherian, and a domain. The graded projective-to-free theorem therefore makes every stalk of \(C\) free. These are the sheaf properties needed by boundary recognition.
Step 4. Prove the identity of C is primitive
The normalized boundary sheaf \(B\) has primitive identity by the indecomposability theorem. Transferring primitivity through the splitting \(i,p\) makes \(e=ip\) primitive.
The adjoint reverses composition, is involutive, and preserves zero. The adjoint-primitivity theorem therefore makes \(e^*\) primitive. Transferring primitivity to its image makes \(1_C\) primitive.
Step 5. Split the normalized top into C
The fixed-stalk theorem gives \(e_b^*=1_{W^b}\). Use the normalized top identifications \(a_0,b_0\) from Step 1, and define \[ a=q_ba_0:B^b\to C^b,\qquad c=b_0j_b:C^b\to B^b. \] Then \[ ca=b_0j_bq_ba_0=b_0e_b^*a_0=b_0a_0=1_{B^b}. \] All these maps preserve degree.
Step 6. Apply boundary recognition
The finite interval, coefficient algebra, quotient rule, free stalks, flabbiness, generation, primitive identity, and top splitting have all been supplied in Steps 1–5.
The boundary-recognition theorem now gives a graded isomorphism \(B\simeq C\). This recognition is the input for proving that the particular canonical comparison map is an isomorphism.
References
The source declarations below give the constructions and proofs.
Lean formalization
OAI.KLInvariance.TitsSpace.bottBoundarySheaf_splitIn: Support.lean:293.
OAI.KLInvariance.TitsSpace.bottSplitAdjointSplit: CharacterReciprocity.lean:1213.
OAI.KLInvariance.TitsSpace.bottSplitAdjoint_free: CharacterReciprocity.lean:1221.
OAI.KLInvariance.TitsSpace.bottSplitAdjointTop_split: CharacterReciprocity.lean:1245.
OAI.KLInvariance.TitsSpace.bottBoundarySplitAdjoint_primitive: BalancedPairings.lean:2025.
OAI.KLInvariance.TitsSpace.bottBoundarySheaf_iso_adjoint: CharacterReciprocity.lean:1258.