Inverse bar coefficients and the R-value kernel identity
For a unit \(q\) and every \(x,y\), the inverse bar transforms give \[ \sum_{z\le y} b_z(q)(x)b_y(q^{-1})(z)=\delta_{x,y}. \] The sum is over the finite lower Bruhat ideal of \(y\). In terms of R-values, this becomes \[ \sum_{z\le y} r_q(x,z)\,q^{\ell(y)}q^{-\ell(z)}r_{q^{-1}}(z,y) =\delta_{x,y}. \]
Proof
Step 1. Expand the inverse transform on one basis vector
The inverse-transform theorem gives \(B_qb_y(q^{-1})=e_y\). Expand the argument in the standard basis: \[ b_y(q^{-1})=\sum_{z\le y}b_y(q^{-1})(z)e_z. \] Triangularity gives its support, and finiteness of the lower ideal makes this a finite sum.
Apply the linear map \(B_q\) and extract the \(x\)-coefficient. This is exactly the first identity.
Step 2. Substitute both R-value normalizations
For one summand, abbreviate \(m=\ell(x)\), \(n=\ell(z)\), \(p=\ell(y)\). The R-value product is \[ \bigl((-1)^m(-q)^n b_z(q)(x)\bigr) q^p q^{-n} \bigl((-1)^n(-q^{-1})^p b_y(q^{-1})(z)\bigr). \] Its sign is \((-1)^{m+2n+p}=(-1)^{m+p}\). Its power of \(q\) is \(q^n q^p q^{-n}q^{-p}=1\). Thus the summand equals \[ (-1)^{\ell(x)+\ell(y)}b_z(q)(x)b_y(q^{-1})(z). \] The remaining scalar is independent of \(z\).
Step 3. Sum and evaluate the diagonal factor
Pull that common scalar out of the finite sum and apply Step 1. If \(x\ne y\), the result is zero. If \(x=y\), the sign is \((-1)^{2\ell(x)}=1\), so the result is one.
This proves the R-value identity. Terms with \(x\not\le z\) vanish by support, so for \(x\le y\) the sum can equivalently be restricted to the interval \([x,y]\).
Step 4. Compute the rank-one cancellation
For \(x=1,y=s\), the two R-value terms are \[ 1\cdot q\cdot(q^{-1}-1)+(q-1)\cdot1=0. \] The two bar-coefficient terms are \[ b_1(q)(1)b_s(q^{-1})(1) +b_s(q)(1)b_s(q^{-1})(s) =(q-1)+(q^{-1}-1)q=0. \] These are the same cancellation under the normalization in Step 2.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.HeckeQ.bar_coeff_inverse: Hecke.lean:1055.
OAI.KLInvariance.HeckeQ.rValue_inverse_factor: Hecke.lean:1069.
OAI.KLInvariance.HeckeQ.rValue_inverse: Hecke.lean:1085.