Theorem Degreewise exact sequences and word-stalk dimensions
✓ theorem proved

Degreewise exact sequences and word-stalk dimensions

Fix a finite-generator Coxeter system and a finite basis of the extended realization, with linear coefficient generators in degree one. For a word \(\mathbf t\), define \[ f_{\mathbf t,x}(n)=\dim_{\mathbb R}(M_{\mathbf t}^x)_n. \] These dimensions are finite.

For every integer \(n\), prepending a simple reflection \(s\) gives \[ f_{s\mathbf t,x}(n)= \begin{cases} f_{\mathbf t,x}(n)+f_{\mathbf t,sx}(n-1),&\ell(x)<\ell(sx),\\ f_{\mathbf t,sx}(n)+f_{\mathbf t,x}(n-1),&\ell(sx)<\ell(x). \end{cases} \]

Relation to the OpenAI notes

Degree-one kernels and shift conventions.

OpenAI notes: The normalization proof temporarily assigns degree two to linear forms and uses \(M(k)^j=M^{j+k}\) for its shifted self-dual object. bg:normalization.

Lean (primary): The word construction keeps linear forms in degree one. Each kernel equivalence sends an old degree-\(n\) vector to degree \(n+1\). The local module-data convention is \(M(1)_n=M_{n-1}\). GradedModules:1244, BottSamelson:1618, BottSamelson:2084, BottSamelson:1744.

Role in the proof: The old degree-\((n-1)\) piece contributes to the new degree-\(n\) kernel. Reindexing its Hilbert sum gives the factor \(q\) in the word-character recurrence.

Proof

Step 1. Check finite-dimensionality of all pieces

A degree-\(n\) coefficient piece is zero for \(n<0\). For \(n\ge0\), it is spanned by monomials in finitely many variables whose exponents sum to \(n\). Each exponent is between \(0\) and \(n\), so there are finitely many such monomials.

A word has finitely many subexpressions. Its weight-coordinate module in a fixed degree is therefore a finite product of finite-dimensional coefficient pieces. The stalk piece is a real subspace of that product, and hence is finite-dimensional. This is the source's coefficient-piece, coordinate-piece and stalk-piece finiteness chain.

Step 2. Restrict a head projection degree by degree

Let \(h:Z\to Q\) be either head projection, viewed as real-linear. It is surjective and preserves degree.

For \(q\in Q_n\), choose \(z\) with \(h(z)=q\), then take the degree-\(n\) component. Homogeneity gives \[ h(\operatorname{pr}_n z) =\operatorname{pr}_n(h(z))=q. \] Thus the restricted map \(Z_n\to Q_n\) is surjective.

Its kernel consists of the vectors in \(Z_n\) killed by \(h\), which is exactly the degree-\(n\) piece of \(\ker h\). Hence \[ 0\to(\ker h)_n\to Z_n\to Q_n\to0 \] is an exact sequence of real vector spaces.

Step 3. Substitute the ascent kernel

Suppose \(\ell(x)<\ell(sx)\). Use \(h_0:Z=M_{s\mathbf t}^x\to X=M_{\mathbf t}^x\). The degree-one kernel theorem identifies \((K_0)_n\simeq(M_{\mathbf t}^{sx})_{n-1}\). Thus Step 2 becomes \[ 0\to(M_{\mathbf t}^{sx})_{n-1} \xrightarrow{\kappa_0}(M_{s\mathbf t}^{x})_n \xrightarrow{h_0}(M_{\mathbf t}^{x})_n\to0. \] Rank-nullity in these finite-dimensional spaces gives the ascent formula.

Step 4. Substitute the descent kernel

Suppose \(\ell(sx)<\ell(x)\). Use the real-linear map \(h_1:Z\to Y=M_{\mathbf t}^{sx}\). Its kernel piece is \((K_1)_n\simeq(M_{\mathbf t}^{x})_{n-1}\), giving \[ 0\to(M_{\mathbf t}^{x})_{n-1} \xrightarrow{\kappa_1}(M_{s\mathbf t}^{x})_n \xrightarrow{h_1}(M_{\mathbf t}^{sx})_n\to0. \] Rank-nullity gives the descent formula.

Simple multiplication changes length by one, so the two cases cover every \(x\). The shift \(n-1\) comes from the actual degree of the root in the kernel map.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.TitsSpace.symmetricGrading_finite: BottSamelson.lean:1731.

OAI.KLInvariance.TitsSpace.bottCoordinatePiece_finite: BottSamelson.lean:1735.

OAI.KLInvariance.TitsSpace.bottStalkPiece_finite: BottSamelson.lean:1740.

OAI.KLInvariance.TitsSpace.bottStalk_finrank_ascent: BottSamelson.lean:1744.

OAI.KLInvariance.TitsSpace.bottStalk_finrank_descent: IntegralPairings.lean:86.

Used by