Theorem The bar eigenvalue of a dihedral lower-interval sum
✓ theorem proved

The bar eigenvalue of a dihedral lower-interval sum

For a rank-two Coxeter system and a unit \(q\) in a commutative ring, \[ B_qS_w=q^{-\ell(w)}S_w. \] Here \(B_q\) is the coefficient-linear unnormalized bar transform.

Relation to the OpenAI notes

The same string calculation in two normalizations.

OpenAI notes: The notes use \(C_w^\circ=q^{-\ell(w)/2}S_w\). Their product has a zero correction at length two and the shorter-prefix vector at longer lengths; induction proves semilinear bar invariance. loc:dihedral-product, loc:dihedral-kl.

Lean (primary): The finite sum S_w has coefficient one below w. The product is S_w plus q times the correction vector. Lean applies the coefficient-linear bar map to the product formula at the inverse unit parameter and cancels the common correction term. Dihedral:217, Dihedral:336, Dihedral:517.

Role in the proof: Writing \(q=v^2\) and \(C_w^\circ=v^{-\ell(w)}S_w\) converts the two product formulas and bar statements directly. The explicit double-string theorem includes the finite longest-element case.

Proof

Step 1. State the reduced-word induction

Prove the formula for the product of every reduced \(\omega_i(m)\), by strong induction on \(m\), allowing either first index.

At \(m=0\), the sum is \(S_1=T_1\), and \(B_qT_1=T_1\).

At \(m=1\), it is \(S_s=T_1+T_s\). The simple bar vector gives \[ B_qS_s=T_1+q^{-1}T_s+(q^{-1}-1)T_1 =q^{-1}(T_1+T_s). \]

Step 2. Supply both smaller induction instances

For \(m=n+2\), use \(w,v,p,C=C_{i,n}\) from the product recurrence. The tail \(v\) has reduced length \(n+1<m\), so induction gives \[ B_qS_v=q^{-(n+1)}S_v. \]

If \(n>0\), the prefix \(p\) is reduced with length \(n<m\), giving \(B_qC=q^{-n}C\). If \(n=0\), then \(C=0\), and this same equation holds because both sides are zero.

Step 3. Apply the linear bar to the inverse-parameter recurrence

Use the product recurrence at parameter \(q^{-1}\): \[ (I+L_s(q^{-1}))S_v=S_w+q^{-1}C. \] Applying the linear map \(B_q\), and using its intertwining identity, gives \[ (I+L_s(q)^{-1})B_qS_v =B_qS_w+q^{-1}B_qC. \] The coefficient \(q^{-1}\) is retained by this linear map.

Step 4. Simplify the left side with the original product recurrence

The quadratic inverse formula gives \[ \begin{aligned} I+L_s(q)^{-1} &=I+q^{-1}\bigl(L_s(q)-(q-1)I\bigr)\\ &=\bigl(1-q^{-1}(q-1)\bigr)I+q^{-1}L_s(q)\\ &=q^{-1}(I+L_s(q)). \end{aligned} \] Here \(1-q^{-1}(q-1)=q^{-1}\), using \(q^{-1}q=1\). Substitute it and Step 2 into the left side: \[ \begin{aligned} (I+L_s(q)^{-1})B_qS_v &=q^{-(n+2)}(I+L_s(q))S_v\\ &=q^{-(n+2)}(S_w+qC)\\ &=q^{-(n+2)}S_w+q^{-(n+1)}C. \end{aligned} \]

The right side of Step 3 becomes \[ B_qS_w+q^{-1}q^{-n}C =B_qS_w+q^{-(n+1)}C. \] Subtract the identical correction term from the two sides. This yields \(B_qS_w=q^{-(n+2)}S_w\), completing induction.

Step 5. Pass from a reduced word to any element

Every element has a reduced alternating expression of length \(\ell(w)\). Apply the proven formula to that expression to obtain the statement for every \(w\).

Step 6. Express the normalized bar-invariant vector

Over Laurent coefficients with \(q=v^2\), put \[ C_w^\circ=v^{-\ell(w)}S_w. \] The coefficient involution sends \(v^{-\ell(w)}\) to \(v^{\ell(w)}\). Since the coefficients of \(S_w\) are integers, its semilinear bar equals \(B_qS_w\). Thus \[ \overline{C_w^\circ} =v^{\ell(w)}v^{-2\ell(w)}S_w =v^{-\ell(w)}S_w=C_w^\circ. \] This is the normalization of the bar calculation used in the OpenAI notes.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.Dihedral.plus_leftInv: Dihedral.lean:509.

OAI.KLInvariance.Dihedral.bar_lowerSum_alt: Dihedral.lean:517.

OAI.KLInvariance.Dihedral.bar_lowerSum: Dihedral.lean:557.

Used by