Definition The Tits form, simple roots, and normal functionals
✓ definition formalized

The Tits form, simple roots, and normal functionals

Let \(I\) be a finite set of simple generators for a Coxeter system \((W,S)\). Write \(m_{ij}\) for the order of \(s_is_j\). Thus \(m_{ii}=1\), and for distinct indices \(m_{ij}\) is an integer at least two or is infinite.

The geometric root space is \(V_0=\mathbb R^I\). For each \(i\), let \(\alpha_i\) be the vector whose \(i\)-th coordinate is one and whose other coordinates are zero. Every vector has a unique expression \(v=\sum_i v_i\alpha_i\).

The defining matrix

Set \[ g_{ij}= \begin{cases} -2\cos(\pi/m_{ij}),&m_{ij}<\infty,\\ -2,&m_{ij}=\infty. \end{cases} \] The Tits form is the bilinear map \[ B:V_0\times V_0\longrightarrow\mathbb R,\qquad B(v,w)=\sum_{i,j\in I}v_i g_{ij}w_j. \] Consequently \(B(\alpha_i,\alpha_j)=g_{ij}\). Symmetry of the Coxeter matrix gives \(g_{ij}=g_{ji}\); interchanging the two summation indices proves \(B(v,w)=B(w,v)\).

The normalization gives \(g_{ii}=-2\cos\pi=2\). For distinct generators, exponent two gives \(g_{ij}=0\), exponent three gives \(g_{ij}=-1\), and an infinite exponent gives \(g_{ij}=-2\). Thus the relations between the generators determine every matrix entry.

Root vectors and normal functionals

The simple normal associated with \(\alpha_i\) is \[ \nu_i:V_0\longrightarrow\mathbb R,\qquad \nu_i(v)=B(\alpha_i,v)=\sum_jg_{ij}v_j. \] A root vector belongs to \(V_0\); its normal belongs to the dual space \(V_0^*=\operatorname{Hom}_{\mathbb R}(V_0,\mathbb R)\). The form defines a linear map \[ B^\flat:V_0\longrightarrow V_0^*,\qquad v\longmapsto B(v,-), \] sending \(\alpha_i\) to \(\nu_i\).

The simple reflection formula is \(s_i(v)=v-\nu_i(v)\alpha_i\). It changes only the \(i\)-th coordinate in the simple-root basis. Since \(\nu_i(\alpha_i)=2\), it sends \(\alpha_i\) to \(-\alpha_i\) and fixes every vector in \(\ker\nu_i\). The involution calculation is developed in Linear reflections from a root and a normal.

Two rank-two calculations

For \(m_{st}=3\), the Gram matrix is \[ \begin{pmatrix}2&-1\\-1&2\end{pmatrix}. \] Its determinant is three, so \(B^\flat\) is invertible. Both the roots and the normals are independent.

For \(m_{st}=\infty\), the matrix is \[ \begin{pmatrix}2&-2\\-2&2\end{pmatrix}. \] Here \(\nu_s(x\alpha_s+y\alpha_t)=2x-2y\) and \(\nu_t(x\alpha_s+y\alpha_t)=-2x+2y\). The roots still form a basis, whereas \(\nu_t=-\nu_s\). The vector \(k=\alpha_s+\alpha_t\) satisfies \(B(k,v)=0\) for every \(v\). The kernel of \(B^\flat\) is called the radical.

The extension construction pairs radical directions with added dual directions. It retains the original roots and gives independent extended normals; see The radical and its dual extension and Nondegeneracy and independence in the extended root space.

References

Realizations in the OpenAI notes.

Lean formalization

OAI.KLInvariance.TitsSpace.gram: Tits.lean:365.

OAI.KLInvariance.TitsSpace.root: Tits.lean:367.

OAI.KLInvariance.TitsSpace.normal: Tits.lean:371.

OAI.KLInvariance.TitsSpace.form: Tits.lean:374.

OAI.KLInvariance.TitsSpace.form_symm: Tits.lean:391.

OAI.KLInvariance.TitsSpace.form_roots: Tits.lean:405.

The Coxeter matrix encodes infinity by zero. In the coefficient formula, real division by zero gives Real.pi / 0 = 0, so the entry is -2 * Real.cos 0 = -2. The diagonal entry is one and gives -2 * Real.cos Real.pi = 2.

Used by