Scaled coordinate lifts and the evaluation denominator
Fix a word \(\mathbf s\) for a finite-generator Coxeter system. Let \(A\) be the symmetric coefficient algebra of its real extended realization, \(M=M_{\mathbf s}\) the word module, and \(J\) its finite subexpression index set. Use the standard word basis to identify \(M\) with \(A^J\). Full evaluation is then represented by a square matrix \[ E:A^J\longrightarrow A^J. \] Define the evaluation denominator \(\Delta=\det E\). Then \(\Delta\ne0\), and every coordinate tuple \(v\in A^J\) has a scaled lift: \[ \exists m\in M,\qquad E(m)=\Delta v. \] In particular, there are elements \(c_\varepsilon\in M\) satisfying \[ E_\eta(c_\varepsilon)= \begin{cases} \Delta,&\eta=\varepsilon,\\ 0,&\eta\ne\varepsilon. \end{cases} \] Each \(c_\varepsilon\) is supported at weight \(w(\varepsilon)\), and every \(m\in M\) satisfies \[ \Delta m=\sum_{\varepsilon\in J}E_\varepsilon(m)c_\varepsilon. \]
Proof
Step 1. Establish the nonzero determinant
The full evaluation matrix is injective over the domain \(A\). Suppose \(\det E=0\). Over the fraction field of \(A\), a square matrix with zero determinant has a nonzero vector \(z\) in its kernel.
The finite list of coordinates of \(z\) has a common nonzero denominator \(a\in A\). Then \(v=az\) has coordinates in \(A\), is nonzero, and satisfies \(Ev=aEz=0\). This contradicts injectivity over \(A\). Thus \(\Delta\ne0\).
Step 2. Construct a lift with the adjugate matrix
The adjugate matrix is the transpose of the matrix of signed minors. Cofactor expansion gives \[ E\,\operatorname{adj}(E)=(\det E)I. \] Its entries lie in \(A\). For \(v\in A^J\), let \(m\) have basis-coordinate vector \(\operatorname{adj}(E)v\). Then \[ E(m)=E\operatorname{adj}(E)v=\Delta v, \] proving the scaled-lift statement.
Step 3. Lift an individual coordinate
Take \(v\) to be the standard coordinate vector at \(\varepsilon\), and choose the resulting \(c_\varepsilon\). Its evaluation is \(\Delta\) at \(\varepsilon\) and zero elsewhere.
If \(w(\eta)\ne w(\varepsilon)\), then \(\eta\ne\varepsilon\), so \(E_\eta(c_\varepsilon)=0\). This is precisely support at \(w(\varepsilon)\).
Step 4. Reconstruct a scaled word element
Evaluate the proposed sum at a fixed coordinate \(\eta\): \[ E_\eta\!\left(\sum_\varepsilon E_\varepsilon(m)c_\varepsilon\right) =\sum_\varepsilon E_\varepsilon(m)E_\eta(c_\varepsilon) =\Delta E_\eta(m). \] This equals \(E_\eta(\Delta m)\). The equality holds at every coordinate, and full evaluation is injective, so the two word elements are equal.
Step 5. Compute the one-letter denominator
In the basis \(1\otimes1,1\otimes\alpha\), \[ E=\begin{pmatrix}1&\alpha\\ 1&-\alpha\end{pmatrix}, \qquad \Delta=-2\alpha. \] The scaled coordinate lifts have evaluation pairs \((-2\alpha,0)\) and \((0,-2\alpha)\). Both are compatible modulo \(\alpha\), and the expansion multiplies any compatible pair \((f,g)\) by \(-2\alpha\).
References
The local source declarations below give the constructions and proofs.
Lean formalization
OAI.KLInvariance.TitsSpace.bottCoordinateDenominator: Frobenius.lean:1317.
OAI.KLInvariance.TitsSpace.bottCoordinateDenominator_ne_zero: Frobenius.lean:1321.
OAI.KLInvariance.TitsSpace.bottEvaluation_contains_scaled_coordinates: Frobenius.lean:1328.
OAI.KLInvariance.TitsSpace.bottCoordinateLift: Frobenius.lean:1448.
OAI.KLInvariance.TitsSpace.bottCoordinateLift_eval: Frobenius.lean:1452.
OAI.KLInvariance.TitsSpace.bottCoordinateLift_supported: Frobenius.lean:1458.
OAI.KLInvariance.TitsSpace.bottCoordinateExpansion: Frobenius.lean:1465.