Theorem Reduced-word weights fill the Bruhat interval
✓ theorem proved

Reduced-word weights fill the Bruhat interval

Let \(\mathbf t\) be a reduced word for \(b\) in a Coxeter system. Then its subexpression weights are exactly the elements below \(b\) in strong Bruhat order: \[ \{w_{\mathbf t}(\eta):\eta\in J_{\mathbf t}\}=[1,b]. \] This weight statement permits an arbitrary simple-generator set.

For a finite simple-generator set, form the actual tensor word module over its real symmetric coefficient algebra. Its weight stalk satisfies \[ M_{\mathbf t}^x\ne0\quad\Longleftrightarrow\quad x\le b. \]

Proof

Step 1. Turn each bit string into a subword

Define \(\nu(\eta)\) by retaining precisely those letters whose bit is \(1\), in their original order. Recursively, \[ \nu(())=[],\qquad \nu(0,\eta)=\nu(\eta),\qquad \nu(1,\eta)=s\,\nu(\eta) \] for a word beginning with \(s\).

At a zero bit the first letter is omitted; at a one bit it is retained. Induction on the word therefore proves that \(\nu(\eta)\) is a sublist of \(\mathbf t\). The product recursion gives \(\operatorname{prod}(\nu(\eta))=w_{\mathbf t}(\eta)\). These are the source's bottSubword_sublist and bottSubword_wordProd.

Step 2. Bound every weight by the reduced-word product

Every word has a reduced subword with the same product. Apply this to \(\nu(\eta)\). The resulting reduced word is still a subword of the fixed reduced word \(\mathbf t\).

The reduced-subword criterion therefore gives \[ w_{\mathbf t}(\eta)=\operatorname{prod}(\nu(\eta))\le b. \] This is the source's bruhat_of_subword application in bottWeight_bruhat.

Step 3. Realize every element of the interval

If \(x\le b\), the reduced-subword criterion supplies a sublist \(\nu\) of \(\mathbf t\) whose product is \(x\). Mark its retained positions by \(1\) and all others by \(0\).

More formally, induction on the sublist witness inserts a zero bit when a letter is skipped and a one bit when it is retained. This constructs \(\eta\) with \(\nu(\eta)=\nu\). Its weight is \(x\), by Step 1. Together with Step 2, this proves the range equality.

Step 4. Translate weight support into stalk support

The stalk \(M_{\mathbf t}^x\) is the image of evaluation into \(A^{J_x}\), where \(J_x=\{\eta:w_{\mathbf t}(\eta)=x\}\).

If \(x\not\le b\), the index set \(J_x\) is empty by Steps 2–3. Its coordinate module is zero, so the stalk is zero.

If \(x\le b\), choose \(\eta\in J_x\). The unit tensor evaluates to \(1\) at this coordinate. Since the real coefficient algebra is nonzero, its stalk projection is a nonzero vector. Thus the stalk is nonzero. This last calculation is the stalk interpretation of the source's exact weight range.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.TitsSpace.bottSubword: Frobenius.lean:1678.

OAI.KLInvariance.TitsSpace.bottSubword_sublist: Frobenius.lean:1682.

OAI.KLInvariance.TitsSpace.bottSubword_wordProd: Frobenius.lean:1693.

OAI.KLInvariance.TitsSpace.exists_bottSubword: Frobenius.lean:1702.

OAI.KLInvariance.TitsSpace.bottWeight_bruhat: Frobenius.lean:1713.

OAI.KLInvariance.TitsSpace.exists_bottWeight: Frobenius.lean:1718.

OAI.KLInvariance.TitsSpace.bottWeight_range: Frobenius.lean:1724.

OAI.KLInvariance.TitsSpace.bottEvaluation_unit: BottSamelson.lean:280.

Used by