Integral injectivity of full word evaluation
Fix a Coxeter system with finite simple-generator set, its real extended realization \(V\), and the symmetric coefficient algebra \(A=\operatorname{Sym}_{\mathbb R}(V)\). For every word \(\mathbf t\), full evaluation is an injective \(A\)-linear map \[ E_{\mathbf t}:M_{\mathbf t}\longrightarrow A^{J_{\mathbf t}}. \] Thus two word vectors are equal whenever all their subexpression evaluations agree.
Proof
Step 1. Express evaluation through basis coordinates
The word basis gives a linear isomorphism \[ c_{\mathbf t}:M_{\mathbf t}\xrightarrow{\sim}A^{J_{\mathbf t}}. \] By the evaluation expansion, \[ E_{\mathbf t}(m)=\mathsf E_{\mathbf t}\,c_{\mathbf t}(m), \] where the matrix \(\mathsf E_{\mathbf t}\) has entry \(E_{\mathbf t,\eta}(b_\gamma)\) in row \(\eta\), column \(\gamma\). We prove injectivity of multiplication by \(\mathsf E_{\mathbf t}\) by induction on the word.
Step 2. Start with the empty word
There is one subexpression. The module is \(A\), its basis is \(1\), and its evaluation is the identity. Thus \(\mathsf E_\varnothing=(1)\), and its multiplication map is injective.
Step 3. Identify the matrix after prepending a letter
Assume multiplication by \(\mathsf E_{\mathbf t}\) is injective. Prepend \(s\), write \(\tau=\chi_s\), and let \(\alpha_s\) be its root polynomial. Splitting the new basis by its first bit gives \[ \mathsf E_{s\mathbf t}= \begin{pmatrix} \mathsf E_{\mathbf t}&\alpha_s\mathsf E_{\mathbf t}\\ \tau(\mathsf E_{\mathbf t})&-\alpha_s\tau(\mathsf E_{\mathbf t}) \end{pmatrix}. \] These four blocks come from the evaluation of \(1\otimes b_\gamma\) and \(1\otimes\alpha_s b_\gamma\) at the two first-bit choices. The sign in the lower-right block is \(\tau(\alpha_s)=-\alpha_s\).
Step 4. Verify the wall-injectivity hypotheses
The symmetric algebra of the real vector space \(V\) is a polynomial algebra after choosing a basis, hence a domain of characteristic zero. The coefficient action \(\tau\) is a ring automorphism. The simple-reflection Frobenius data give \(\alpha_s\ne0\).
The wall-injectivity theorem therefore applies to the matrix from Step 3 and the induction hypothesis. It proves injectivity of \(\mathsf E_{s\mathbf t}\). Since the tail word is shorter by one letter, Steps 2–4 establish the assertion for every finite word.
Step 5. Return to word vectors
If \(E_{\mathbf t}(m)=E_{\mathbf t}(n)\), Step 1 and matrix injectivity give \(c_{\mathbf t}(m)=c_{\mathbf t}(n)\). The coordinate map is an isomorphism, so \(m=n\).
For one letter, this recovery is the calculation \[ (f,g)=(u+\alpha_s v,u-\alpha_s v). \] Equal outputs force equality of \(2\alpha_s v\), then of \(v\), and then of \(u\). The induction repeats this recovery with the old evaluation matrix in each block.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.TitsSpace.bottEvaluation_expand: Frobenius.lean:1216.
OAI.KLInvariance.TitsSpace.bottEvaluationMatrix_cons: Frobenius.lean:1225.
OAI.KLInvariance.TitsSpace.bottEvaluationMatrix_mulVec_cons: Frobenius.lean:1247.
OAI.KLInvariance.TitsSpace.bottEvaluationMatrix_injective: Frobenius.lean:1263.
OAI.KLInvariance.TitsSpace.bottEvaluation_injective: Frobenius.lean:1290.