The word pairing lowers degree by the word length
Let \(A\) be the graded symmetric coefficient algebra of a finite-generator real extended realization. For any word \(\mathbf s\) of length \(m\), let \(M_{\mathbf s}\) carry its tensor-defined perfect pairing and word grading. Then \[ \langle(M_{\mathbf s})_i,(M_{\mathbf s})_j\rangle \subseteq A_{i+j-m}. \] Thus each prepended simple reflection increases the pairing shift by one.
Proof
Step 1. Compute the degree of a reflection trace
For a simple reflection, write \(\tau\) for its action on \(A\) and \(\alpha\in A_1\) for its nonzero root polynomial. The reflection decomposition is \[ a=a_0+\alpha a_1,\qquad a_0,a_1\in A^\tau, \] and the trace extracts \(a_1\). Applying \(\tau\) gives \(\tau(a)=a_0-\alpha a_1\), hence \[ 2\alpha\,\operatorname{tr}(a)=a-\tau(a). \]
If \(a\in A_n\), the action preserves degree, so \((a-\tau(a))/2\in A_n\). It equals \(\alpha\,\operatorname{tr}(a)\). Multiplication by \(\alpha\) is injective in the coefficient domain. The homogeneous-scalar cancellation theorem therefore gives \[ \operatorname{tr}(a)\in A_{n-1}. \]
Step 2. Start the induction with the empty word
The empty-word module is \(A\), with basis vector \(1\) in degree zero and pairing given by multiplication. Its basis pairing is \(\langle1,1\rangle=1\in A_0\), agreeing with shift zero.
We prove by induction that for word-basis vectors \(b_\varepsilon,b_\eta\), \[ \langle b_\varepsilon,b_\eta\rangle \in A_{d_\varepsilon+d_\eta-m}, \] where \(d_\varepsilon\) is the number of selected \(1\)-choices in the tensor basis.
Step 3. Prepend one letter and calculate on its basis
Suppose the assertion holds for a word \(\mathbf t\) of length \(m\). Prepend \(s\). The reflection basis over \(A^s\) is \(r_0=1,r_1=\alpha_s\), of degrees \(0,1\). The induced word basis is \[ b_{(a,\varepsilon)}=1\otimes r_a b_\varepsilon, \qquad a\in\{0,1\}, \] of degree \(a+d_\varepsilon\).
The tensor-pairing formula gives \[ \langle b_{(a,\varepsilon)},b_{(c,\eta)}\rangle =\operatorname{tr}_s\!\left( r_a r_c\langle b_\varepsilon,b_\eta\rangle\right). \] By induction, the expression inside the trace has degree \(a+c+d_\varepsilon+d_\eta-m\). Step 1 lowers this degree by one, giving \[ (a+d_\varepsilon)+(c+d_\eta)-(m+1). \] This is exactly the required basis identity for the longer word.
Step 4. Extend the basis identity to homogeneous vectors
Let \(u=\sum_\varepsilon a_\varepsilon b_\varepsilon\in(M_{\mathbf s})_i\) and \(v=\sum_\eta c_\eta b_\eta\in(M_{\mathbf s})_j\). The homogeneous-coordinate theorem gives \[ a_\varepsilon\in A_{i-d_\varepsilon},\qquad c_\eta\in A_{j-d_\eta}. \] Bilinearity expands the pairing into terms \[ a_\varepsilon c_\eta\langle b_\varepsilon,b_\eta\rangle. \] Each has degree \[ (i-d_\varepsilon)+(j-d_\eta) +(d_\varepsilon+d_\eta-m)=i+j-m. \] The finite sum therefore lies in \(A_{i+j-m}\), proving the theorem.
Step 5. Recover the one-letter pairing
For one reflection, the basis \(1\otimes1,1\otimes\alpha\) has degrees \(0,1\). Its pairing matrix is \[ \begin{pmatrix}0&1\\ 1&0\end{pmatrix}, \] because the trace of \(1\) and \(\alpha^2\) is zero and the trace of \(\alpha\) is one. The nonzero entries pair degrees \(0,1\) into degree \(0+1-1=0\).
For a reduced word with product \(b\), its length is \(\ell(b)\). This supplies the shift used by the boundary pairing and the stalk–costalk reciprocity formula.
References
The source declarations below give the constructions and proofs.
Lean formalization
OAI.KLInvariance.ReflectionFrobenius.trace_difference: Frobenius.lean:456.
OAI.KLInvariance.ReflectionFrobenius.trace_homogeneous: Frobenius.lean:464.
OAI.KLInvariance.FrobeniusDuality.inducedPairing_basis: Frobenius.lean:580.
OAI.KLInvariance.FrobeniusDuality.inducedPairing_basis_homogeneous: Frobenius.lean:602.
OAI.KLInvariance.TitsSpace.bottBasis_pairing_homogeneous: Frobenius.lean:979.
OAI.KLInvariance.TitsSpace.bottDuality_homogeneous: Frobenius.lean:1024.