Theorem The descent recursion for normalized R-polynomials
✓ theorem proved

The descent recursion for normalized R-polynomials

If \(s\) is a simple reflection and \(sy<y\), then \[ \widetilde R_{x,y}(a)= \begin{cases} \widetilde R_{sx,sy}(a),&sx<x,\\ \widetilde R_{sx,sy}(a)+a\widetilde R_{x,sy}(a),&sx>x. \end{cases} \] Together with diagonal value \(1\) and vanishing outside Bruhat order, this computes the family by induction on the upper endpoint's length. The rank-one value is \(\widetilde R_{1,s}(a)=a\).

Relation to the OpenAI notes

Same polynomial, different construction order.

OpenAI notes: The path-formula theorem introduces the normalized R-polynomial through increasing reflection paths and relates it to the ordinary R-polynomial; decreasing paths follow by reversing the reflection order. bg:paths, eq:bg-path-formula.

Lean (primary): The normalized R-family is defined first as a coefficient of the Hecke bar transform. Its recursion is proved algebraically, and the geometric path product is later proved equal to that independently constructed family. Hecke:482, Hecke:491, ReflectionOrders:1229.

Role in the proof: The bar construction supplies the R-family and its recursion. The path theorem then identifies that family with the ordered path polynomial.

Proof

Step 1. Relate the two upper endpoints

Fix a simple reflection \(s\) with \(sy<y\), and put \(v=sy\). Then \(sv=y\) is an ascent from \(v\). Write \(B_v\) for the bar-basis vector at \(v\), with the parameter convention \(-a\) used to define \(\widetilde R\).

The bar-basis ascent formula gives \(B_y=L_s^{-1}B_v\). The quadratic identity says that the inverse operator is \(L_s-(-a)=L_s+a\) in this parameter convention.

Step 2. Read the coefficient of a general vector

Write \(B_v=\sum_w b_we_w\). The basis formula with parameter \(-a\) is \[ L_se_w=e_{sw}-a\mathbf1_{\ell(sw)<\ell(w)}e_w. \] The coefficient of \(e_x\) in \(L_sB_v\) is \(b_{sx}-a\mathbf1_{\ell(sx)<\ell(x)}b_x\): the first term comes from \(w=sx\), and the second from \(w=x\).

Adding \(aB_v\), the coefficient in \(L_s^{-1}B_v\) is therefore \[ b_{sx}+a\bigl(1-\mathbf1_{\ell(sx)<\ell(x)}\bigr)b_x. \]

Step 3. Substitute the R-coefficients

By definition, \(b_{sx}=\widetilde R_{sx,sy}(a)\) and \(b_x=\widetilde R_{x,sy}(a)\). If \(sx<x\), the indicator is one and the second term cancels. If \(sx>x\), it is zero and the second term is \(a\widetilde R_{x,sy}(a)\). These are exactly the two branches in the statement.

Since the upper endpoint \(sy\) is shorter than \(y\), this recurrence is suitable for induction. At \(x=y\) the diagonal coefficient is one; outside Bruhat order triangularity gives zero, so these cases are consistent with the same constructed family.

References

Related paper context.

Lean formalization

OAI.KLInvariance.Hecke.rTilde_recursion: Hecke.lean:491.

Used by