Theorem Nondegeneracy and independence in the extended root space
✓ theorem proved

Nondegeneracy and independence in the extended root space

Let \(I\) be finite, let \(V_0=\mathbb R^I\) carry its Tits form \(B\), and let \(K\) be the radical. Choose a linear projection \(p:V_0\to K\) that fixes every element of \(K\). On \(E=V_0\oplus K^*\), use \[ B_p((v,\phi),(w,\psi)) =B(v,w)+\phi(pw)+\psi(pv). \] Then \(B_p\) is symmetric and nondegenerate and restricts to \(B\) on \(V_0\oplus0\). Both families \[ \widetilde\alpha_i=(\alpha_i,0),\qquad \widetilde\nu_i=B_p(\widetilde\alpha_i,-) \] are linearly independent. Finally, \[ \dim E=|I|+\dim K. \] These conclusions also hold for the extended form chosen from this construction in the Lean development.

Proof

Step 1. Produce the projection

Choose a vector-space complement \(U\) of \(K\). Write each \(v\in V_0\) uniquely as \(u+k\) with \(u\in U\), \(k\in K\), and set \(p(v)=k\). Uniqueness of the decomposition proves linearity. If \(v=k\in K\), its decomposition is \(0+k\), so \(p(k)=k\).

Step 2. Check symmetry and restriction

Each summand in the displayed definition is bilinear. Symmetry of \(B\) gives \[ \begin{aligned} B_p((w,\psi),(v,\phi)) &=B(w,v)+\psi(pv)+\phi(pw)\\ &=B(v,w)+\phi(pw)+\psi(pv)\\ &=B_p((v,\phi),(w,\psi)). \end{aligned} \] For two embedded vectors \((v,0),(w,0)\), the two evaluation terms vanish. Hence their pairing is \(B(v,w)\).

Step 3. Test orthogonality against the added dual directions

Suppose \(x=(v,\phi)\) satisfies \(B_p(x,y)=0\) for every \(y\in E\). Taking \(y=(0,\psi)\) gives \[ 0=B_p((v,\phi),(0,\psi))=\psi(pv) \] for every \(\psi\in K^*\). Linear functionals separate vectors, so \(pv=0\).

To see the separation explicitly, if \(pv\ne0\), extend that vector to a basis of \(K\) and define a functional taking value one on it. This contradicts the displayed equality.

Step 4. Test against radical vectors

For \(k\in K\), symmetry gives \(B(v,k)=B(k,v)=0\). Also \(p(k)=k\) and \(pv=0\). Taking \(y=(k,0)\) therefore gives \[ 0=B_p((v,\phi),(k,0))=0+\phi(k)+0. \] Thus \(\phi\) vanishes on all of \(K\), so \(\phi=0\).

Step 5. Test against all original vectors

Taking \(y=(w,0)\), with arbitrary \(w\in V_0\), now gives \(B(v,w)=0\). Hence \(v\in K\). The projection fixes radical vectors, so \(pv=v\). Step 3 showed \(pv=0\), and therefore \(v=0\).

We have proved that a vector orthogonal to every vector in the second argument is zero. Symmetry gives the same assertion for the other argument. This proves nondegeneracy.

Step 6. Prove root and normal independence

Suppose \(\sum_i c_i\widetilde\alpha_i=0\). Projecting to \(V_0\) gives \(\sum_i c_i\alpha_i=0\). Since the \(\alpha_i\) form a basis, every coefficient is zero.

Now suppose \(\sum_i c_i\widetilde\nu_i=0\). Bilinearity rewrites this equation as \[ B_p\left(\sum_i c_i\widetilde\alpha_i,-\right)=0. \] Nondegeneracy forces \(\sum_i c_i\widetilde\alpha_i=0\), and root independence forces all \(c_i=0\). Thus the normals are independent as well.

Step 7. Compute the dimension

The simple-root basis has \(|I|\) elements. A basis of the finite-dimensional radical \(K\) and its dual basis have the same size. Consequently \[ \dim E=\dim V_0+\dim K^*=|I|+\dim K. \] The dimension-minimality theorem shows that this number is attained by the smallest possible nondegenerate extension preserving the embedded form.

References

2. Real realizations and moment-graph sheaves.

Lean formalization

OAI.KLInvariance.TitsSpace.extendedForm_spec: Tits.lean:558.

OAI.KLInvariance.TitsSpace.extendedRoot_linearIndependent: Tits.lean:569.

OAI.KLInvariance.TitsSpace.extendedNormal_linearIndependent: Tits.lean:575.

OAI.KLInvariance.TitsSpace.extended_finrank: Tits.lean:613.

Used by