Uniqueness of a normalized polynomial-family pair
For a fixed Coxeter system, any two pairs satisfying the normalization data have identical R-components and identical P-components.
Relation to the OpenAI notes
Explicit construction of the normalized family.
OpenAI notes: The conventions specify the ordinary R recurrence, the strict P degree bound and reciprocity, referring to the classical Kazhdan–Lusztig normalization. sec:introduction, eq:reciprocity.
Lean (primary): After constructing R, the development defines P by the negative lower half of the proper-upper defect. The reversed R-kernel proves that defect anti-reciprocal, yielding reciprocity. Separate uniqueness inductions identify the selected family with the construction. CanonicalKL:70, CanonicalKL:162, Bruhat:204, CanonicalKL:260, CanonicalKL:282.
Role in the proof: The KB presents the coefficient solution, defect identity, rank induction and seven normalization fields as a continuous construction. The classical selection interfaces are then identified with the explicit formulas.
Proof
Step 1. Induct on the upper length for the R-component
Let \((R,P)\) and \((R',P')\) be normalized pairs. Prove \(R_{x,y}=R'_{x,y}\) by strong induction on \(\ell(y)\), for all \(x\).
If \(y=1\), then \(x=1\) gives diagonal value one. If \(x\ne1\), it cannot lie below \(1\): length monotonicity would force \(\ell(x)=0\), hence \(x=1\). Both R-values are therefore zero.
For \(y\ne1\), choose a simple left descent \(s\), and put \(v=sy\), with \(\ell(v)<\ell(y)\). Both families satisfy the same recurrence. In the descent branch at \(x\), induction identifies \(R_{sx,v}\) with \(R'_{sx,v}\). In the ascent branch it identifies both \(R_{x,v}\) and \(R_{sx,v}\) with their primed counterparts. The scalar coefficients \(q-1,q\) are the same. Hence the R-values agree for every pair.
Step 2. Set the independent rank induction for P
Now prove \(P_{x,y}=P'_{x,y}\) by strong induction on the natural rank difference, allowing every pair.
If \(x\not\le y\), both values are zero. If \(x=y\), both are one.
For \(x<y\), set \(d=d(x,y)\). If \(z\in[x,y]\) and \(z\ne x\), then \[ d(z,y)=\ell(y)-\ell(z)<\ell(y)-\ell(x)=d. \] Thus induction gives \(P_{z,y}=P'_{z,y}\) at every index except possibly the bottom.
Step 3. Subtract the two reciprocity equations
Set \(f=P_{x,y}-P'_{x,y}\). Linearity of reflection and Step 1 give \[ \operatorname{reflect}_df =\sum_{z\in[x,y]}R_{x,z}(P_{z,y}-P'_{z,y}). \] For \(z\ne x\), Step 2 makes the bracket zero. At \(z=x\), the R-factor is \(R_{x,x}=1\). Therefore \[ \operatorname{reflect}_df=f. \]
Step 4. Apply low-degree rigidity
Both P-values have twice their natural degree strictly less than \(d\). The degree of their difference is at most the maximum of their degrees, so \(2\,\operatorname{natDegree}(f)<d\).
The integral low-degree rigidity theorem now gives \(f=0\), hence \(P_{x,y}=P'_{x,y}\). This completes the second induction.
Step 5. Recover equality of pairs
Step 1 proves equality of the first component as a function of both endpoints. Steps 2–4 prove equality of the second. Equality of the two ordered pairs follows componentwise.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.normalized_R_unique: Bruhat.lean:111.
OAI.KLInvariance.normalized_P_unique: Bruhat.lean:151.
OAI.KLInvariance.normalizedKL_unique: Bruhat.lean:204.