Definition The two head-projection kernels and their graded pieces
✓ definition formalized

The two head-projection kernels and their graded pieces

Fix a finite-generator Coxeter system, a finite basis of its extended realization, and the symmetric coefficient algebra \(A\) with its degree-one grading. Fix a word \(\mathbf t\), a simple reflection \(s\), and \(x\in W\). Put \[ X=M_{\mathbf t}^x,\qquad Y=M_{\mathbf t}^{sx},\qquad Z=M_{s\mathbf t}^x,\qquad \tau=\chi_s,\quad \alpha=\alpha_s. \]

For \(y\in Y\), write \([y]_\tau\) for the same underlying vector in \({}^\tau Y\). Its scalar law is \[ a[y]_\tau=[\tau(a)y]_\tau. \] The twist has the same real-vector-space structure because \(\tau\) fixes real scalars.

The reflected pair lattice \(L_x\subseteq X\oplus{}^\tau Y\) has an \(A\)-linear coordinate isomorphism \(C_x:L_x\to Z\). Its two head projections are \[ h_0(C_x(p,[q]_\tau))=p,\qquad h_1(C_x(p,[q]_\tau))=q. \] The first map is \(A\)-linear. The second is \(\tau\)-semilinear over \(A\) and real-linear, with its value viewed in the underlying \(Y\). Both preserve degree.

Both maps are surjective: to attain a prescribed value in \(X\) or \(Y\), choose a tail-word lift of that value, take its two stalk projections as a lattice pair, and apply \(C_x\).

The two kernels

Define \[ K_0=\ker h_0,\qquad K_1=\ker h_1. \] The source records \(K_0\) as the kernel of the \(A\)-linear first projection and \(K_1\) as the kernel of the real-linear second projection. We use their real-vector-space gradings \[ (K_i)_n=\{z\in K_i:z\in Z_n\}. \] These are bottHeadKernelPiece and bottHeadRightKernelPiece.

Since \(h_i\) preserves degree, it commutes with homogeneous components. If \(h_i(z)=0\), then \(h_i(z_n)=0\) for every component \(z_n\). Thus each kernel is closed under components, and these pieces give its direct-sum grading.

The ascent kernel map

If \(\ell(x)<\ell(sx)\), exact reflection kernels for the tail identify the first lattice-projection kernel with pairs \((0,[\alpha y]_\tau)\), \(y\in Y\). Transporting through \(C_x\) gives \[ \kappa_0:Y\xrightarrow{\sim}K_0,\qquad y\longmapsto C_x(0,[\alpha y]_\tau), \] as a real-linear equivalence. The map \([y]_\tau\mapsto C_x(0,[\alpha y]_\tau)\) is \(A\)-linear on the twisted copy of \(Y\); the real-linear identification with the underlying \(Y\) gives the displayed map.

This is bottHeadKernelEquiv. Injectivity uses the nonzero root acting on the stalk coordinates, and surjectivity is the exact lattice-kernel statement.

The descent kernel map

If \(\ell(sx)<\ell(x)\), use the upward tail edge \(sx\to x\). The second lattice-projection kernel consists of \((\alpha z,0)\), \(z\in X\). Therefore \[ \kappa_1:X\xrightarrow{\sim}K_1,\qquad z\longmapsto C_x(\alpha z,0) \] is a real-linear equivalence. It is bottHeadRightKernelEquiv.

Transporting the kernel through coordinate isomorphisms

For a linear map \(f:L\to P\) and an isomorphism \(e:L\to Z\), the map \(u\mapsto e(u)\) identifies \[ \ker f\ \simeq\ \ker(f e^{-1}). \] The inverse is \(e^{-1}\). If the target is also changed by an injective linear isomorphism \(g:P\to Q\), the condition \(g(f(e^{-1}z))=0\) is equivalent to \(f(e^{-1}z)=0\).

The source names these constructions kernelSourceEquiv, kernelTargetEquiv and kernelMixedEquiv. In the descent case, the last construction combines the \(A\)-linear lattice isomorphism with the real-linear map forgetting the scalar twist.

References

The local source declarations below give the construction and proof.

Lean formalization

OAI.KLInvariance.kernelSourceEquiv: BottSamelson.lean:1551.

OAI.KLInvariance.kernelTargetEquiv: BottSamelson.lean:1971.

OAI.KLInvariance.kernelMixedEquiv: BottSamelson.lean:1999.

OAI.KLInvariance.TitsSpace.bottHeadKernelEquiv: BottSamelson.lean:1578.

OAI.KLInvariance.TitsSpace.bottHeadKernelPiece: BottSamelson.lean:1597.

OAI.KLInvariance.TitsSpace.bottHeadRight_surjective: BottSamelson.lean:2031.

OAI.KLInvariance.TitsSpace.bottDescentKernel: BottSamelson.lean:2037.

OAI.KLInvariance.TitsSpace.bottHeadRightKernelEquiv: BottSamelson.lean:2045.

OAI.KLInvariance.TitsSpace.bottHeadRightKernelPiece: BottSamelson.lean:2067.

Used by