Theorem The reciprocal convolution identity for R-polynomials
✓ theorem proved

The reciprocal convolution identity for R-polynomials

For a polynomial \(p(q)=\sum_{i=0}^d c_iq^i\) of degree at most \(d\), its reflection in degree \(d\) is \[ \operatorname{reflect}_d p=\sum_{i=0}^d c_iq^{d-i}=q^dp(q^{-1}). \] For example, reflecting \(2+3q\) in degree two gives \(2q^2+3q\). It is a linear operation on polynomials of degree at most \(d\), and reflecting twice returns the original coefficients.

Write \(d(x,z)=\ell(z)-\ell(x)\) for comparable endpoints and \[ \overline R_{x,z}(q)=q^{d(x,z)}R_{x,z}(q^{-1}). \] For every finite Bruhat interval, \[ \sum_{x\leq z\leq y}\overline R_{x,z}(q)R_{z,y}(q)=\delta_{x,y}. \] Terms with noncomparable endpoints vanish. The degree bound on \(R\) ensures that the displayed reflection is a polynomial.

Proof

Step 1. Work with an invertible formal parameter

Let \(Q\) be the Laurent indeterminate, so \(Q\) is a unit in \(\mathbb Z[Q,Q^{-1}]\). The constructed ordinary R-polynomial evaluated at \(Q\) equals the coefficient R-value of the unnormalized Hecke bar transform.

The inverse bar-coefficient theorem substitutes the factors \((-1)^{\ell(x)}(-Q)^{\ell(z)}\) and \((-1)^{\ell(z)}(-Q^{-1})^{\ell(y)}\) into the coefficient sum. Its calculation cancels all powers at the intermediate vertex and leaves a diagonal sign equal to one. It gives \[ \sum_{x\le z\le y} R_{x,z}(Q)\,Q^{\ell(y)}Q^{-\ell(z)} R_{z,y}(Q^{-1})=\delta_{x,y}. \] This is the \((x,y)\)-entry of the identity operator obtained by composing the bar transforms at \(Q\) and \(Q^{-1}\). Triangularity removes terms outside \([x,y]\), so the coefficient sum is finite.

Step 2. Return to polynomial coefficients

Since \(z\le y\), \(Q^{\ell(y)}Q^{-\ell(z)}=Q^{d(z,y)}\). The ordinary R-degree bound gives \(\deg R_{z,y}\le d(z,y)\). Consequently \(Q^{d(z,y)}R_{z,y}(Q^{-1})\) is the Laurent image of the polynomial \(\operatorname{reflect}_{d(z,y)}R_{z,y}\).

Every term of Step 1 is therefore the image of a polynomial. The embedding \(\mathbb Z[q]\to\mathbb Z[Q,Q^{-1}]\) is injective, so \[ \sum_{x\le z\le y}R_{x,z}\operatorname{reflect}_{d(z,y)}R_{z,y} =\delta_{x,y} \] holds in the polynomial ring.

Step 3. Reverse the order of the reflected factor

Reflect this identity in total degree \(d(x,y)\). For each comparable \(z\), \(d(x,z)+d(z,y)=d(x,y)\). Using the degree bounds, reflection of the product splits as \[ \operatorname{reflect}_{d(x,y)} \bigl(R_{x,z}\operatorname{reflect}_{d(z,y)}R_{z,y}\bigr) =\operatorname{reflect}_{d(x,z)}R_{x,z}\,R_{z,y}. \] Here reflecting the second polynomial twice returns it; this follows term by term from \(q^i\mapsto q^{d(z,y)-i}\).

Reflection distributes over the finite sum. If \(x\ne y\), the right-hand side is zero and stays zero. If \(x=y\), the rank is zero and reflection fixes one. This gives the stated reversed convolution.

For noncomparable \(x,y\), each potential term vanishes by triangularity, and the right-hand side is zero as well.

References

Related paper context.

Lean formalization

OAI.KLInvariance.OrdinaryR.polynomial_kernel_reverse: OrdinaryR.lean:245.

OAI.KLInvariance.OrdinaryR.polynomial_kernel: OrdinaryR.lean:190.

OAI.KLInvariance.eval₂_reflect_unit: Hecke.lean:1101.

Used by