Definition The pair-projection sheaf of a Bott–Samelson module
✓ definition formalized

The pair-projection sheaf of a Bott–Samelson module

Fix a finite-generator Coxeter system, its real extended realization \(V\), and a Bruhat interval \([u,b]\). Let \(M_{\mathbf s}\) be a Bott–Samelson word module over \(A=\operatorname{Sym}_{\mathbb R}(V)\), and let \(\pi_x:M_{\mathbf s}\twoheadrightarrow M_{\mathbf s}^x\) be its weight-stalk projections. On an edge \(e:x\to y\), form the pair-projection quotient \[ B^e=(M_{\mathbf s}^x\oplus M_{\mathbf s}^y)/ \operatorname{im}(\pi_x,\pi_y). \] The lower map sends \(v\) to \([(v,0)]\), and the upper map sends \(w\) to \(-[(0,w)]\). The pair-projection calculation gives \[ \rho^-_e(v)=\rho^+_e(w) \quad\Longleftrightarrow\quad \exists m\in M_{\mathbf s},\ (\pi_x(m),\pi_y(m))=(v,w). \] Thus edge compatibility expresses simultaneous lifting of the two endpoint values.

For one reflection, the full evaluation image is \(\{(f,g):f-g\in\alpha A\}\). Both stalk images are \(A\), and the quotient above is \(A/(\alpha)\), via \([(f,g)]\mapsto f-g\bmod\alpha\). The lower and upper maps both become reduction modulo \(\alpha\).

Sections supplied by word elements

Every \(m\in M_{\mathbf s}\) gives a global section \((\pi_x(m))_x\), because it lifts its own two projections on every edge. Since each stalk projection is onto its image, these global sections generate every stalk.

For a reduced word on its full interval \([1,b]\), the global-section realization theorem proves that the section map \(\Psi\) is an isomorphism. A sheaf endomorphism \(F\) acts on sections through its vertex maps; conjugation gives the word-module endomorphism \[ T=\Psi^{-1}\Gamma(F)\Psi. \] The descent construction below describes the induced stalk and edge maps.

The word-sheaf flabbiness theorem extends compatible assignments from upper subsets of the interval. Together with free stalks, exact upper quotients and the normalized top stalk, it gives the word-boundary summand theorem.

Graded maps and pair quotients

The evaluation projections preserve degree, and their image stalks inherit the image grading. Each pair relation is a homogeneous image, so its edge quotient has the quotient grading. Both restriction maps have degree zero.

An endomorphism of the word module preserving the kernels of both endpoint projections descends to their stalks. The quotient-descent theorem then supplies an edge endomorphism satisfying both commuting squares. This construction is used to turn the integral word adjoint into a graded sheaf endomorphism.

The actual finite graded module data

Choose a finite basis of \(V\) and use the associated degree-one grading of \(A\). The source packages the constructions as follows.

Construction Module or map
bottWordData The tensor word module with its word-basis grading.
bottStalkData The evaluation image stalk with its coordinatewise image grading.
bottStalkHom The degree-zero surjection from the word module to the stalk.
bottEdgeData The graded quotient of the two endpoint stalks by their simultaneous image.
bottLowerHom and bottUpperHom The degree-zero lower and upper maps into that quotient.
bottGradedProjectionSheaf These vertex modules, edge modules and restrictions, assembled on the interval.

The word module is finitely generated by its finite tensor basis. Its stalks are finite images, and the edge quotient is a quotient of the direct sum of two finite stalks.

The gradings are nonnegative. For a negative-degree word vector, every evaluation lies in a negative coefficient piece, hence is zero; evaluation injectivity forces the word vector to be zero. For a negative-degree stalk vector, every coordinate is zero. The pair quotient inherits nonnegativity from the two stalks.

These facts supply the finite graded module data used at every vertex and edge.

Polynomial coordinates

The chosen basis induces a graded isomorphism \[ e:A\xrightarrow{\sim}P,\qquad P=\mathbb R[X_i:i\text{ indexes the basis}]. \] The polynomial word sheaf transports each vertex and edge module through \(e\), and transports the restriction maps by their underlying functions. Its \(P\)-action on an old vector is \(p\cdot m=e^{-1}(p)m\). This is bottPolynomialSheaf.

Write \(S_P\) for this polynomial sheaf. A symmetric root label \(\alpha_E\) becomes the linear coordinate polynomial \(e(\alpha_E)\). The polynomial word-stalk theorem proves that this transport preserves stalk freeness and gives the exact upper edge kernel \(e(\alpha_E)S_P^y\).

References

Related paper context.

Lean formalization

OAI.KLInvariance.TitsSpace.bottProjectionSheaf: Frobenius.lean:1761.

OAI.KLInvariance.PairProjection.relation: Decomposition.lean:153.

OAI.KLInvariance.PairProjection.compatible_iff: Decomposition.lean:168.

OAI.KLInvariance.TitsSpace.bottWordData: BottSamelson.lean:1449.

OAI.KLInvariance.TitsSpace.bottStalkData: BottSamelson.lean:1427.

OAI.KLInvariance.TitsSpace.bottStalkHom: BottSamelson.lean:1459.

OAI.KLInvariance.TitsSpace.bottEdgeData: BottSamelson.lean:1468.

OAI.KLInvariance.TitsSpace.bottLowerHom: BottSamelson.lean:1472.

OAI.KLInvariance.TitsSpace.bottUpperHom: BottSamelson.lean:1477.

OAI.KLInvariance.TitsSpace.bottGradedProjectionSheaf: BottSamelson.lean:1482.

OAI.KLInvariance.TitsSpace.bottPolynomialSheaf: Support.lean:417.

Used by