A reduced tensor word realizes all global sections
Fix a finite-generator Coxeter system and a reduced word \(\mathbf t\) with product \(b\). Let \(S\) be its pair-projection sheaf on \(I=[1,b]\), over the symmetric coefficient algebra \(A\).
The section map \[ \Psi:M_{\mathbf t}\longrightarrow\Gamma(S),\qquad \Psi(m)=(\pi_x(m))_{x\in I} \] is an \(A\)-linear isomorphism. Thus every compatible global assignment is realized by a unique word vector.
Relation to the OpenAI notes
Explicit word-image construction of the extension property.
OpenAI notes: The moment-graph sheaf theorem gives extension of compatible sections on upper sets for the normalized boundary sheaf through the cited Braden–MacPherson and Soergel bridges. bg:bmp.
Lean (primary): The word module gives coordinate images and stable coefficient characters. Reflection geometry makes each prime congruence class a single weight or a reflection pair. Explicit diagonal operators combine class lifts, and prime-denominator removal proves compatible-tuple lifting from image projectivity. The word induction supplies upper-image projectivity using its two kernel sequences. IntegralPairings:470, IntegralPairings:467, IntegralPairings:1299, IntegralPairings:1408, IntegralPairings:2057, IntegralPairings:1620, IntegralPairings:1833, BalancedPairings:292, Support:293.
Role in the proof: Upper-image lifting gives interval flabbiness after extension to the ambient upper closure. Full-interval section realization also has a direct free-image proof. Free stalks, root quotients, flabbiness, generation and the normalized word top then supply the graded boundary summand.
Proof
Step 1. Prove injectivity of the full vertex projection
Let \[ E=\prod_{x\in I}M_{\mathbf t}^x,\qquad v:M_{\mathbf t}\to E,\quad v(m)=(\pi_x(m))_x, \] and put \(L=\operatorname{im}v\).
Every subexpression weight of the reduced word lies in \(I\). If \(v(m)=0\), the stalk at each such weight is zero on \(m\), so every subexpression evaluation of \(m\) vanishes. Full evaluation is injective, giving \(m=0\).
Thus \(v\) identifies the finite free word module with \(L\), making \(L\) free. The same argument proves injectivity of \(\Psi\).
Step 2. Supply a nonzero scalar lift of any vertex assignment
For a tuple \(z\in E\), define \[ c_\varepsilon=(z_{w(\varepsilon)})_\varepsilon. \] This is defined for every subexpression because its weight lies in \(I\). Scaled coordinate lifting gives a word vector whose evaluations are \(\Delta_{\mathbf t}c_\varepsilon\). Its vertex projections are therefore \(\Delta_{\mathbf t}z_x\) at every \(x\), so \[ \Delta_{\mathbf t}z\in L,\qquad \Delta_{\mathbf t}\ne0. \]
Step 3. Supply a prime-local scalar lift of a section
Now suppose \(z\) is a compatible global section, and fix a prime \(p\). The interval \(I\) is finite. Its coefficient weights \(x\mapsto\chi_x(a)\) stabilize \(L\), since right multiplication on the word module has those projection scalars.
A singleton residue class lifts by stalk surjectivity. A class with two distinct weights is a reflection pair by the prime-character theorem. The two lengths differ, so compatibility on the corresponding upward interval edge gives one word vector realizing both class values.
The class-lift assembly theorem therefore supplies \(d_p\) with \[ p\nmid d_p,\qquad d_pz\in L. \] This is the source's bottSection_local_denominator argument on the finite interval itself.
Step 4. Remove the scalar multipliers
The coefficient algebra is a polynomial UFD over \(\mathbb R\). Each stalk, and therefore \(E\), is torsion-free. Step 1 makes \(L\) a free submodule of \(E\); Step 2 gives its generic multiplier, and Step 3 gives the multiplier avoiding each prime.
Apply the free-target prime-denominator membership theorem. It gives \(z\in L\), so some \(m\in M_{\mathbf t}\) has \(v(m)=z\). Because both tuples are compatible, this is precisely \(\Psi(m)=z\). Thus \(\Psi\) is surjective.
Step 5. Obtain the isomorphism and its one-letter form
The section map is linear, injective and surjective, so it is the isomorphism bottSectionEquiv.
For one letter, it sends the basis coordinates \(u,v\) of \(u(1\otimes1)+v(1\otimes\alpha)\) to \[ (f,g)=(u+\alpha v,u-\alpha v). \] A compatible pair has \(f-g\in\alpha A\), and its inverse coordinates are \[ u=(f+g)/2,\qquad v=(f-g)/(2\alpha). \] The divisibility condition makes the second coordinate an element of \(A\). This is the explicit instance of the general section-realization isomorphism.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.TitsSpace.bottVertexProjection: BalancedPairings.lean:174.
OAI.KLInvariance.TitsSpace.bottVertexProjection_injective: BalancedPairings.lean:179.
OAI.KLInvariance.TitsSpace.bottVertexProjection_stable: BalancedPairings.lean:185.
OAI.KLInvariance.TitsSpace.bottVertexProjection_scaled: BalancedPairings.lean:194.
OAI.KLInvariance.TitsSpace.bottSection_pair_lift: BalancedPairings.lean:212.
OAI.KLInvariance.TitsSpace.bottSection_prime_lift: BalancedPairings.lean:231.
OAI.KLInvariance.TitsSpace.bottSection_local_denominator: BalancedPairings.lean:262.
OAI.KLInvariance.TitsSpace.bottSection_surjective: BalancedPairings.lean:277.
OAI.KLInvariance.TitsSpace.bottSectionEquiv: BalancedPairings.lean:292.