Theorem Simultaneous decomposition of Bott–Samelson stalk and kernel characters
✓ theorem proved

Simultaneous decomposition of Bott–Samelson stalk and kernel characters

Fix a reduced word \(\mathbf s\) with product \(b\) in a finite-generator Coxeter system. Choose a real basis of the extended realization and use the corresponding polynomial coefficient algebra, with root labels in degree one. There exist polynomials \(\mu_y(q)\in\mathbb Z[q]\), \(y\in[1,b]\), with nonnegative coefficients, such that for every \(x\in[1,b]\), \[ C_x^{\mathbf s}(q) =\sum_{x\le y\le b}\mu_y(q)C_{x,y}(q), \] \[ K_x^{\mathbf s}(q) =\sum_{x\le y\le b}\mu_y(q)\, q^{\ell(y)-\ell(x)}C_{x,y}(q^{-1}). \] Here \(C_x^{\mathbf s}\) is the word-stalk character and \(K_x^{\mathbf s}\) the full outgoing-kernel character of its word sheaf. Moreover \(\mu_b(q)=1\).

Proof

Step 1. Check the word-sheaf and model inputs

The polynomial-coordinate word sheaf has free stalks and the upper-quotient rule by the word-stalk and edge-kernel theorems. Its word sections generate each stalk, and the reduced-word flabbiness theorem gives flabbiness.

The word section pairing is perfect and support-orthogonal, so it supplies section duality. Together with the free-stalk, quotient, and generation data, the costalk theorem identifies each full outgoing kernel with the dual of its finite free stalk. This proves the required kernel freeness. The actual lower-supported boundary models have free stalks by construction and free full outgoing kernels by the lower-model kernel theorem. All coefficients, labels, and graph vertices are those of the common finite interval \([1,b]\).

Step 2. Decompose the sheaf and record its shifts

Apply the model-decomposition theorem to obtain a graded sheaf isomorphism \[ B^{\mathbf s}\simeq\bigoplus_{y,j}O_y(d_{y,j}), \qquad d_{y,j}\in\mathbb N. \] Define \[ \mu_y(q)=\sum_jq^{d_{y,j}}. \] The sums are finite, and every coefficient counts occurrences of a nonnegative shift.

Step 3. Substitute the model stalk characters

The simultaneous model-character theorem gives \[ C_x^{\mathbf s}(q)=\sum_y\mu_y(q)C_{O_y}(x;q). \] For \(x\le y\), the model stalk is the actual boundary stalk on \([1,y]\), with character \(C_{x,y}\). For \(x\not\le y\), the model stalk is zero. Removing those zero terms yields the first displayed formula.

Step 4. Substitute the model kernel characters

The same theorem and the same shifts give \[ K_x^{\mathbf s}(q)=\sum_y\mu_y(q)K_{O_y}(x;q). \] For \(x\le y\), the lower-model kernel theorem identifies the term as \(q^{\ell(y)-\ell(x)}C_{x,y}(q^{-1})\). For the other centers it is zero. This gives the second formula.

Each reflected expression is a polynomial because its exponents are the nonnegative degrees of the model's kernel basis.

Step 5. Determine the top multiplicity

The reduced-word top-stalk theorem gives \(C_b^{\mathbf s}=1\). Among centers \(y\le b\), the support condition permits a nonzero stalk at \(b\) only when \(b\le y\), hence only when \(y=b\). That model has normalized top character \(C_{b,b}=1\).

The first formula at \(x=b\) therefore gives \[ 1=\mu_b(q)\cdot1. \] Thus \(\mu_b=1\), the coefficient used when lower-center terms are cancelled in actual-character reciprocity.

References

The local source declarations below give the constructions and proofs.

Lean formalization

OAI.KLInvariance.TitsSpace.bottCharacter_decomposition_both: CharacterReciprocity.lean:2019.

Used by