Theorem Bruhat support and diagonal coefficients of the bar basis
✓ theorem proved

Bruhat support and diagonal coefficients of the bar basis

Write \(b_y(x)\) for the coefficient of \(e_x\) in \(b_y\). In both Hecke conventions, \[ b_y(x)=0\qquad\text{if }x\not\le y. \] The diagonal values are \[ b_y(a)(y)=1,\qquad b_y(q)(y)=q^{-\ell(y)}. \]

Proof

Step 1. Extract coefficients of inverse operators

For \(f=\sum_w f(w)e_w\), only \(w=sx\) and \(w=x\) can contribute to the \(x\)-coefficient after applying a simple inverse.

In the normalized convention, its basis formula gives \[ (L_s(a)^{-1}f)(x)= \begin{cases} f(sx),&sx<x,\\ f(sx)-af(x),&sx>x. \end{cases} \] The coefficient of \(f(sx)\) is one in both cases. The subtraction of \(af(x)\) remains only when \(x\) is an ascent.

In the unnormalized convention, the inverse basis formula is \[ L_s(q)^{-1}T_w= \begin{cases} T_{sw},&sw<w,\\ q^{-1}T_{sw}+(q^{-1}-1)T_w,&sw>w. \end{cases} \] Substituting \(w=sx\) reverses the ascent/descent condition. Consequently \[ (L_s(q)^{-1}f)(x)= \begin{cases} q^{-1}f(sx),&sx<x,\\ f(sx)+(q^{-1}-1)f(x),&sx>x. \end{cases} \]

Step 2. Locate both possible source coefficients in Bruhat order

Let \(y=sv\) with \(v<sv\). If \(x\not\le y\), then \(x\not\le v\), since \(v\le y\).

Also \(sx\not\le v\). Indeed, if \(sx\le v\), monotonicity of the upper simple envelope would give \[ U_s(sx)\le U_s(v)=sv=y. \] Since \(U_s(sx)=U_s(x)\) and \(x\le U_s(x)\), this would imply \(x\le y\), a contradiction.

Thus both \(x\) and \(sx\) lie outside the lower ideal of \(v\).

Step 3. Prove triangular support by induction

Induct on a reduced word for \(y\), with the assertion for every \(x\). For \(y=1\), the vector is \(e_1\), so every coefficient outside \(x=1\) is zero.

For a reduced word \(s\omega\) with tail product \(v\), the ascent formula gives \(b_y=L_s^{-1}b_v\). If \(x\not\le y\), Step 2 and the induction hypothesis give \[ b_v(x)=b_v(sx)=0. \] Every branch in Step 1 is therefore zero. This proves support in \(\{x:x\le y\}\).

Step 4. Compute the normalized diagonal

Use the same reduced-tail induction. At \(y=1\), the diagonal coefficient is one.

At \(y=sv\), we have \(sy=v<y\). The normalized coefficient formula in Step 1 therefore gives \[ b_y(a)(y)=b_v(a)(v)=1 \] by induction.

Step 5. Compute the unnormalized diagonal

Again the base value is one. At \(y=sv\), Step 1 gives \[ b_y(q)(y)=q^{-1}b_v(q)(v). \] Induction makes this \(q^{-1}q^{-\ell(v)}=q^{-\ell(y)}\), because \(\ell(y)=\ell(v)+1\).

For the two-element lower interval, the full vector is \(b_s(q)=q^{-1}T_s+(q^{-1}-1)T_1\): its support and diagonal exhibit both assertions directly.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.Hecke.leftInv_coeff: Hecke.lean:424.

OAI.KLInvariance.Hecke.simple_mul_le_ascent: Hecke.lean:431.

OAI.KLInvariance.Hecke.barBasis_zero: Hecke.lean:441.

OAI.KLInvariance.Hecke.barBasis_diagonal: Hecke.lean:466.

OAI.KLInvariance.HeckeQ.leftInv_coeff: Hecke.lean:895.

OAI.KLInvariance.HeckeQ.barBasis_zero: Hecke.lean:941.

OAI.KLInvariance.HeckeQ.barBasis_diagonal: Hecke.lean:966.

Used by