Prerequisites: quotient spaces and duality and bilinear forms and inertia.
Orthogonality for a symmetric bilinear form is an equation about pairings. It need not describe positive lengths, and a subspace can meet its own orthogonal complement in a nonzero line. The radical explains how dimensions change in the degenerate case, while the inertia explains how large a totally isotropic subspace can be. This lesson connects those two pieces of bilinear-form geometry.
Radicals, complements and isotropic subspaces
Throughout, is a symmetric bilinear form on a finite-dimensional real vector space, with .
Definition 10.1.1
(Radical and orthogonal complement)
The radical of is
For a subspace , its orthogonal complement is
The form is nondegenerate when .
The radical is contained in every orthogonal complement. The radical of the restriction of to is a different object.
Definition 10.1.2
(Restricted radical and totally isotropic subspaces)
The radical of the restriction to is
A nonzero vector is isotropic if . A subspace is totally isotropic if for every .
A vector in must pair to zero with vectors of , not necessarily with every vector of . Thus the restricted radical can be larger than .
Common mistake
(The restricted radical is not )
It is tempting to say that is degenerate only when it contains radical vectors of . On take the hyperbolic form , whose matrix has determinant , so . For we have , so although . A restriction can be degenerate inside a nondegenerate form.
Proposition 10.1.3
(Total isotropy is tested on single vectors)
A subspace is totally isotropic if and only if for every .
Proof
(Polarization)
If is totally isotropic, then on by taking . Conversely, if for all , then for the vector lies in , and symmetry gives
Hence .
Common mistake
(One isotropic vector does not make a plane totally isotropic)
On with , the vectors and are both isotropic, since and . But , so the plane they span is not totally isotropic. Pairings between different vectors still matter.
Recovering the missing dimension through the radical
Theorem 10.1.4
(Dimension of the orthogonal complement)
For every subspace ,
When is nondegenerate this reads .
Proof
(Nondegenerate case, then the quotient by the radical)
First suppose is nondegenerate. Then the map sending to is an isomorphism. Every functional on extends to by extending a basis, so the map
is surjective. Its kernel is , and rank–nullity gives .
For a possibly degenerate form, pass to . The rule is well defined because changing either representative adds zero pairings. It is nondegenerate: if a class pairs to zero with every class, its representative lies in .
The image has dimension . A vector lies in exactly when its quotient class lies in . Applying the nondegenerate dimension formula in the quotient, where , and adding the dimension of the kernel of the quotient map gives
The extra term is essential: radical directions in impose no equations on the vectors perpendicular to .
Proof X-ray
What the radical correction records
The proof has two layers. The nondegenerate case uses exactly one fact about : the map is an isomorphism, which makes the restriction map surjective, so rank–nullity applies without a defect. The second layer is a reduction, not a new computation: the quotient is built precisely so that the degenerate directions are removed, and the check that is well defined and nondegenerate is where the definition of is used both ways. The final count has three terms: , the shrinkage of to by , and the kernel added back. The first and last cancel, leaving the single correction . That term is essential: a radical direction in imposes no equation on the vectors perpendicular to , so such a direction does not reduce the size of .
Taking the orthogonal complement twice
Theorem 10.1.5
(Double orthogonal complement)
For every subspace ,
Proof
(Inclusion plus equal dimensions)
Symmetry gives , and radical vectors also belong to this double complement. Thus . To prove equality, apply the dimension formula again. Since ,
The inclusion and equal dimensions prove the identity.
The familiar identity with alone is recovered when , or more generally when . Taking two complements in a degenerate ambient form cannot distinguish from the larger subspace obtained by adjoining the radical.
Counterexample mode
Taking the complement twice returns U
On let , so , and let . Then , and . This is two-dimensional while is one-dimensional, so . It equals , as the theorem predicts. The statement is therefore valid only when , in particular for nondegenerate forms.
Quick check
(In the form of the previous example, let . Compute and .)
Find first.
Solution · Answer
Here and , so and . Directly, , so . Then .
Quick check
(Suppose . What is ?)
Use the double complement theorem.
Solution · Answer
, because . The ordinary double-complement identity holds whenever the subspace already contains the radical.
When an orthogonal direct sum exists
Theorem 10.1.6
(Orthogonal direct sum criterion)
if and only if is nondegenerate, that is, .
Proof
(Zero intersection and a dimension count)
If , the intersection is zero. Conversely, if , then as well, since . The dimension formula gives , so the two spaces have total dimension and zero intersection. They form the claimed direct sum.
This criterion can hold even when the ambient form is degenerate, because its radical then belongs to the other summand. It can fail even when the ambient form is nondegenerate, because the restriction to may have a nonzero radical.
There is a constructive version. Let be a basis of a nondegenerate and let be its Gram matrix. To find the component of in the decomposition (so that ), solve
Invertibility of gives the unique coefficients. The resulting projection is self-adjoint relative to because the two summands are mutually perpendicular. It need not be symmetric in the Euclidean coordinate metric. An indefinite form does not provide a positive squared-distance function, so this algebraic decomposition has no automatic nearest-point interpretation.
Worked example 10.1.7
(A degenerate ambient form with a direct sum)
Take on , with . Then , and . The restriction to has Gram matrix , which is nondegenerate, so even though has a nonzero ambient radical. As a check of the projection formula, take : then and , so the -component is and . The example distinguishes ambient degeneracy from degeneracy of the restricted form.
Worked example 10.1.8
(A nondegenerate form with a degenerate restriction)
On take and . The Gram matrix is , with determinant , so . Since , we get . The dimension formula is satisfied: . But , so : the sum is only the line . Here is itself a totally isotropic line, and .
Inertia bounds a totally isotropic subspace
Write the real inertia as : positive, negative and zero counts. In suitable coordinates,
and the remaining coordinates describe the radical.
Theorem 10.1.9
(Largest totally isotropic subspace)
If has real inertia , then every totally isotropic subspace has , and this bound is attained:
Proof
(Two injective projections, then the radical, then an attaining subspace)
First consider the nondegenerate part. On a totally isotropic subspace, projection onto the positive coordinates is injective: if , the equation forces . Projection onto the negative coordinates is injective for the same reason. Thus a totally isotropic subspace there has dimension at most .
In the degenerate space, the image of in is totally isotropic for the induced form, and has dimension
Since , this proves . The bound is attained: take the radical together with the vectors for , where and are matching positive and negative coordinate vectors. Every pair of those chosen vectors has zero pairing, since and the radical pairs to zero with everything.
Concept lensGeometric
Isotropic subspaces as graphs of isometries
In the nondegenerate part, a totally isotropic subspace has a concrete shape. Because the projection to the positive coordinates is injective, it is the graph of a linear map defined on a subspace . Total isotropy says , so preserves lengths, and by polarization it preserves inner products. Conversely, the graph of any such length-preserving is totally isotropic. The bound is then the statement that a length-preserving map is injective, so and .
An isotropic vector is not necessarily radical. On a nonzero totally isotropic subspace, the restricted form is zero and degenerate; the criterion for orthogonal direct sums therefore rules out . The zero subspace is the harmless exception, since its restricted form is nondegenerate in dimension zero.
Worked example 10.1.10
(Maximal isotropic subspaces, including the semidefinite edge case)
On let , so the inertia is and . The theorem gives the maximum dimension . The subspace attains it: and pairs to zero with everything. No totally isotropic subspace has dimension , by the theorem. The semidefinite edge case is on , with : the bound is , and the only totally isotropic lines are inside the radical , since forces .
Quick check
(A symmetric form on has inertia . What is the largest dimension of a totally isotropic subspace?)
Apply the theorem to the three counts.
Solution · Answer
The maximum is . An attaining subspace is spanned by the radical vector together with and , where are the positive and negative coordinate vectors.
Summary of results
The following statements hold for real symmetric bilinear forms; alternating forms require their own dimension bounds, and polarization by two does not apply in characteristic two.
- , which need not equal .
- .
- .
- exactly when the restricted form is nondegenerate.
- For real inertia , the largest totally isotropic dimension is .