Theorem The minimal dimension of a nondegenerate extension
✓ theorem proved

The minimal dimension of a nondegenerate extension

Let \(V_0,E\) be finite-dimensional real vector spaces and \(B\) a bilinear form on \(V_0\). Put \[ K=\{v\in V_0:B(v,w)=0\text{ for every }w\in V_0\}. \] Let \(C\) be a nondegenerate bilinear form on \(E\). Suppose an injective linear map \(\iota:V_0\to E\) satisfies \[ C(\iota(v),\iota(w))=B(v,w). \] Then \[ \dim E\ge\dim V_0+\dim K. \] For symmetric \(B\), the dual extension \(V_0\oplus K^*\) attains this bound.

Relation to the OpenAI notes

Formulation of dimension minimality.

OpenAI notes: The realization lemma gives the lower bound for a nondegenerate extension and also discusses minimal dimension under independent-root and independent-coroot conditions. bg:realization.

Lean (primary): The local dimension theorem assumes an injective form-preserving map into a finite-dimensional space with a nondegenerate bilinear form. It proves that the ambient dimension is at least the original dimension plus its radical dimension. Tits:522.

Role in the proof: The theorem here states this form-preserving extension bound. The constructed dual extension attains it and supplies independent embedded roots and normals.

Proof

Step 1. Locate the embedded subspaces

Set \(U=\iota(V_0)\), \(L=\iota(K)\). Injectivity gives \(\dim U=\dim V_0\) and \(\dim L=\dim K\). For \(k\in K\), \(v\in V_0\), \[ C(\iota(k),\iota(v))=B(k,v)=0. \] Define \(L^\perp=\{x\in E:C(l,x)=0\text{ for every }l\in L\}\). The displayed equality proves \(U\subseteq L^\perp\).

Step 2. Compute the orthogonal-complement dimension

Consider \[ \Phi:E\longrightarrow L^*,\qquad \Phi(x)(l)=C(l,x). \] Its kernel is \(L^\perp\). We show it is surjective.

Take \(\eta\in L^*\). Extend a basis of \(L\) to one of \(E\), and assign value zero to the added basis vectors. This extends \(\eta\) to \(\widehat\eta\in E^*\).

Nondegeneracy makes \(x\mapsto C(-,x)\) an injective map \(E\to E^*\). Both spaces have the same finite dimension, so the map is surjective. Choose \(x\) representing \(\widehat\eta\). Restricting to \(L\) gives \(\Phi(x)=\eta\).

Rank–nullity therefore gives \[ \dim L^\perp=\dim E-\dim L. \]

Step 3. Deduce the lower bound

Using \(U\subseteq L^\perp\), \[ \dim V_0=\dim U\le\dim L^\perp=\dim E-\dim K. \] Rearrange to obtain the assertion. This is the orthogonal-subspace argument of the local Lean proof.

Step 4. Attain the bound

For symmetric \(B\), the dual-extension theorem gives a nondegenerate form on \(V_0\oplus K^*\) preserving \(B\). A basis of \(K\) and its dual basis have the same number of elements. Thus \[ \dim(V_0\oplus K^*)=\dim V_0+\dim K, \] as required.

References

Realizations in the OpenAI notes.

Lean formalization

OAI.KLInvariance.Realization.extension_dimension_minimal: Tits.lean:522.

OAI.KLInvariance.Realization.extension_finrank: Tits.lean:517.