Theorem Polynomiality, specialization and descent of ordinary R-values
✓ theorem proved

Polynomiality, specialization and descent of ordinary R-values

The Laurent R-value \(r_Q(x,y)\) is the image of an integral polynomial. Its polynomial representative \(R_{x,y}\) has diagonal one, vanishes outside Bruhat order, and satisfies the ordinary descent recurrence for every left descent of \(y\).

For every unit \(q_0\) in any commutative ring, \[ R_{x,y}(q_0)=r_{q_0}(x,y). \]

Relation to the OpenAI notes

Algebraic construction behind the polynomial conventions.

OpenAI notes: The conventions specify the ordinary R descent recurrence. The reflection-order path theorem introduces the normalized family through path counting and states its square-parameter relation to ordinary R. sec:introduction, bg:paths.

Lean (primary): The regular operators are built from Coxeter length. Left/right commutation and the cyclic identity vector prove reduced-word independence. Inverse operators define R-values; their recurrence proves polynomiality, and a second matching-recurrence argument proves signed parameter reciprocity. Hecke:705, Hecke:1000, OrdinaryR:31, Dihedral:440.

Role in the proof: The KB develops these algebraic constructions before the geometric path formula. The recurrence and inverse-parameter coefficient identities used by the character argument follow from the constructed operators.

Proof

Step 1. Set the polynomiality induction

Induct strongly on \(\ell(y)\), with the statement for every lower label \(x\). We seek a polynomial \(p\) whose Laurent image is \(r_Q(x,y)\).

When \(y=1\), use \(p=1\) if \(x=1\). If \(x\ne1\), then \(x\not\le1\): Bruhat comparison would imply \(\ell(x)\le0\), hence \(x=1\). The support theorem gives \(r_Q(x,1)=0\), so use \(p=0\).

Step 2. Construct the polynomial in a descent step

For \(y\ne1\), a reduced word supplies a left descent \(s\). Put \(v=sy\), so \(\ell(v)<\ell(y)\).

Induction supplies polynomials \(p,r\) whose Laurent images are \(r_Q(x,v)\) and \(r_Q(sx,v)\). If \(sx<x\), the R-value recurrence identifies \(r_Q(x,y)\) with the image of \(r\).

If \(sx>x\), the recurrence identifies it with the image of \[ (q-1)p(q)+q\,r(q). \] This expression uses only nonnegative powers. Thus both branches provide the required polynomial and complete the induction.

Step 3. Identify the source's truncation definition

If \(r_Q(x,y)=p(Q)\), truncating its Laurent series to nonnegative exponents returns \(p\), coefficient by coefficient. Therefore the source's definition has \[ R_{x,y}(Q)=r_Q(x,y). \] This equality also shows that the representative is unique.

Step 4. Transfer initial values and every descent recurrence

The Laurent image of \(R_{x,x}\) is one, so injectivity of the polynomial inclusion gives \(R_{x,x}=1\). The support case similarly gives \(R_{x,y}=0\).

For any left descent \(s\), apply the R-value recurrence at \(Q\). Replace every R-value by its polynomial image. Inclusion preserves addition and multiplication, so injectivity gives exactly the displayed polynomial recurrence.

Step 5. Prove specialization by a second induction

Fix \(q_0\in C^\times\). Induct on \(\ell(y)\), again for all \(x\). At \(y=1\), the diagonal/support cases give the same value on both sides.

For \(y\ne1\), choose \(s\) with \(sy<y\). The polynomial recurrence evaluates to \[ R_{x,y}(q_0)= \begin{cases} R_{sx,sy}(q_0),&sx<x,\\ (q_0-1)R_{x,sy}(q_0)+q_0R_{sx,sy}(q_0),&sx>x. \end{cases} \] Use induction at \((sx,sy)\) and \((x,sy)\). The right-hand side becomes the R-value recurrence at \(q_0\), which equals \(r_{q_0}(x,y)\). This proves specialization.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.OrdinaryR.rValue_polynomial: OrdinaryR.lean:31.

OAI.KLInvariance.OrdinaryR.toLaurent_polynomial: OrdinaryR.lean:65.

OAI.KLInvariance.OrdinaryR.polynomial_diagonal: OrdinaryR.lean:70.

OAI.KLInvariance.OrdinaryR.polynomial_zero: OrdinaryR.lean:75.

OAI.KLInvariance.OrdinaryR.polynomial_recursion: OrdinaryR.lean:80.

OAI.KLInvariance.OrdinaryR.polynomial_eval: OrdinaryR.lean:93.

Used by