Georg Schöckle Universität Konstanz

Fachbereich Mathematik und Statistik 9.12.2020

Skew-symmetric bilinear forms

Proseminar Linear Algebra WS 20/21
Georg Schöckle

Abstract:
This talk should give an outline to the properties of skew-symmetric bilinear forms. We will
explore the equivalent representation matrix and investigate its properties. Further we will
investigate in the subspaces that can be spanned by the skew-symmetric bilinear forms and
the orthogonality of these spaces. At the end we will address the creation of skew-symmetric
bilinear forms from linear forms.

1
 Georg Schöckle

1 Definitions:
1.1 Fields, Spaces
Let K be a field, let V, U, W be vector spaces over that field.

1.2 Bilinearity
Let α, α1 , α2 ∈ U and β, β1 , β2 ∈ V . A bilinear map is a function

f :U ×V →W

such that

(i) f (α1 + cα2 , β) = f (α1 , β) + cf (α2 , β)

(ii) f (α, β1 + cβ2 ) = f (α, β1 ) + cf (α, β2 )

A bilinear form is a bilinear map, with f : V × V → K.

1.3 Symmetric bilinear form

Let α, β ∈ V . A Symmetric bilinear form is a bilinear form, with:

f (α, β) = f (β, α)

1.4 Alternating bilinear form

Let α ∈ V . A, alternating bilinear form is a bilinear form with:

f (α, α) = 0

1.5 Skew-symmetric bilinear form

Let α, β ∈ V . A skew-symmetric bilinear form is a bilinear form with:

f (α, β) = −f (β, α)

2
 Georg Schöckle

2 Skew-symmetric forms
2.1 Skew-symmetric ⇔ symmetric if and only if char(K) = 2
Let f be a skew-symmetric bilinear form f is symmetric if and only if char(K) = 2.
Proof: Let f be any skew-symmetric bilinear form over a vector space V and a field K. This
implies
f (α, β) = −f (α, β) α, β ∈ V
it follows either a) or b) because of the characteristic

(a)
f (α, β) = 0

(b)
f (α, β) = 1

The form is in both cases skew-symmetric. In case (a) it is trivial and in case (b) it holds
1 = −1. The same argument is true for the opposite direction.

2.2 Skew-symmetric ⇔ alternating if and only if char(K) 6= 2

Let f be a skew-symmetric bilinear form f is alternating if and only if char(K) 6= 2.
Proof: Let f be any skew-symmetric bilinear form over a vector space V and a field K. This
implies
f (α, α) = −f (α, α) = 0 ∀α ∈ V
So f is alternating.

Suppose now f is a alternating. α, β ∈ V .

0 = f (α + β, α + β) − f (β, β) − f (α, α)
| {z } | {z } | {z }
0 0 0
= f (α, α) + f (α, β) + f (β, α) + f (β, β) − f (β, β) − f (α, α)
= f (α, β) + f (β, α)

It follows f (α, β) = −f (β, α) for any α, β ∈ V , so f is skew-symmetric.

3
 Georg Schöckle

3 Representation Matrix
3.1 Definition representation matrix
Let B = {x1 , . . . , xn } be our ordered basis of V , s be any bilinear form.

([s]B )i,j := s(xi , xj )

3.2 Definition skew-symmetric matrices

The matrix A is skew-symmetric if and only if

AT = −A

Remark
The representation matrix of a bilinear form s is skew-symmetric if and only if s is skew-
symmetric.
([s]B )ij := s(xi , xj ) = −s(xj , xi ) = − ([s]B )ji

3.3 Remark on determinants

Let A be a n × n skew-symmetric matrix. Then:

det(A) = det(AT ) = det(−A) = (−1)n det(A)

So if n is odd the determinant is 0 unless the underlying field has characteristic 2. We will
see what happens when n is even later.

4
 Georg Schöckle

4 Subspaces
4.1 Subspace of the skew-symmetric bilinear forms
The set of all skew-symmetric bilinear forms creates a subspace of L(V, V, K) on any vector
space V and field K.
Proof: We know that all bilinear forms define a vector space L2 (V )
We only need to show closure with scalar multiplication and addition with itself.
Let f, g be skew-symmetric bilinear forms. α, β ∈ V

(f + g)(α, β) = f (α, β) + g(α, β) = −f (β, α) − g(β, α) = −(f + g)(β, α)

Decomposition of this space

L(V, V, K) is the direct sum of the subspace of skew-symmetric bilinear forms (ssL(V, V, K))
and the symmetric bilinear forms (sL(V, V, K)).
Proof: Show ssL(V, V, K) ∩ sL(V, V, K) = {0}. Let f ∈ ssL(V, V, K) ∩ sL(V, V, K). It holds
for α, β ∈ V :
f (α, β) = f (β, α) = −f (α, β) = −f (β, α)
This is always true if char(K) = 2. In any other case it implies f = 0. Now choose a
f ∈ L(V, V, K) and show that there exists g ∈ sL(V, V, K) and h ∈ ssL(V, V, K). For
α, β ∈ V let:
1
g(α, β) = (f (α, β) + f (β, α))
2
1
h(α, β) = (f (α, β) − f (β, α))
2
it follows:
1 1
g(α, β) = (f (α, β) + f (β, α)) = (f (β, α) + f (α, β)) = g(β, α)
2 2
1 1
h(α, β) = (f (α, β) − f (β, α)) = − (f (β, α) − f (α, β)) = −h(β, α)
2 2
and
1 1
f (α, β) = (f (α, β) + f (β, α)) + (f (α, β) − f (β, α)) = g(α, β) + h(α, β)
2 2
So
L(V, V, K) = sL(V, V, K) ⊕ ssL(V, V, K)

4.2 Orthogonal decomposition lemma

Let f be a fixed non zero skew-symmetric bilinear form, let α, β ∈ V be vectors such f (α, β) =
1 (It is clear that α and β are linearly independent). Let γ ∈ span(α, β), with γ = cα + dβ,
c, d ∈ K.
f (γ, α) = f (cα + dβ, α) = cf (α, α) + df (β, α) = −d
f (γ, β) = f (cα + dβ, β) = cf (α, β) + df (β, β) = c
so
γ = f (γ, β)α − f (γ, α)β
Because of the linear independence of α and β if γ = 0 it follows that c = d = 0 so
f (γ, β) = f (γ, α) = 0.
Let W = span(α, β) and let W ⊥ = {x ∈ V |∀γ ∈ W : f (x, γ) = 0} we claim that V = W ⊕W ⊥ .

5
 Georg Schöckle

Choose y ∈ W ∩ W ⊥ for all γ ∈ W applies f (y, γ) = 0 but there is also e, f ∈ K with

y = eα + gβ, if we use γ like above it follows:

f (y, γ) = f (eα + gβ, cα + dβ) = ecf (α, α) + edf (α, β) + gcf (β, α) + gdf (β, β) = ec − gd = 0

So ec = gd for all c, d ∈ K. So e = g = 0 and therefore y = 0 so {0} = W ∩ W ⊥

Now show that any vector x ∈ V has the form x = δ + γ, with δ ∈ W ⊥ and γ ∈ W .
Let γ = f (x, β)α − f (x, α)β and δ = ε − x.
γ ∈ W is clear. For δ holds:

f (δ, α) = f (x − f (x, β)α + f (x, α)β, α)

= f (x, α) + f (x, α)f (β, α)
=0

f (δ, β) = f (x − f (x, β)α + f (x, α)β, β)

= f (x, β) − f (x, β)f (α, β)
=0

Thus δ ∈ W ⊥ . Therefore V = W ⊕ W ⊥ .

4.3 Theorem on further decomposition

The finite dimensional vector space V has a finite number of orthogonal subspaces W0 , . . . , Wk
k ∈ N with V = W1 ⊕ · · · ⊕ Wk ⊕ W0 .
Proof:
We will use W , W ⊥ and f from above and investigate the restriction of f to W ⊥ .
This restriction could be the zero form. If not we can find vectors α1 , β1 ∈ W ⊥ with
f (α0 , βn0 ) = 1. Let these two vectors span 0
o another subspace W and let the be complement
⊥ 0 0
W0 = δ ∈ W | f (α , δ) = f (β , δ) = 0 .

V = W ⊕ W 0 ⊕ W0

If f is still not the zero form on W0 we can divide W0 further. If we repeat we get an finite
number of pairs:
(α1 , β1 ), (α2 , β2 ), . . . , (αk , βk )
with their related spans W1 , . . . , Wk for k ∈ N.

V = W1 ⊕ · · · ⊕ Wk ⊕ W0

Remark
The decomposition depends on f so k differs to other forms. Therefore the decomposition of
V is not unique. But the decomposition may also change if we choose other α1 , β1 .

4.4 Representation matrix of skew-symmetric bilinear forms

There exists a basis of V such that the representation matrix of f relative to this base has
the form:
diag {S, S, . . . , S, 0, . . . , 0}
were !
0 1
S=
−1 0

6
 Georg Schöckle

Proof: If f is again the zero form the result is trivial. In any other case we use the vectors
α1 , β1 , · · · αk , βk that span the orthogonal subspaces we determent above. These vectors are
all linearly independent because:

1. if αi , βi were dependent f (αi , βi ) = 0

2. if αi were dependent to the other vectors the sum of the space in which αi is in wouldn’t
be direct.

We now add the vectors γ1 , . . . , γs which are a basis of W0 so

(α1 , β1 , · · · , αk , βk , γ1 , · · · , γs )

is a basis for V . Now the representation matrix of f to this basis looks like this

··· ···
 
f (α1 , α1 ) f (α1 , β1 ) f (α1 , α2 ) f (α1 , β2 ) 0 0 0 0
 f (β , α )
 1 1 f (β1 , β1 ) f (β1 , α2 ) f (β1 , β2 ) ··· 0 0 0 ··· 0 

f (α2 , α1 )

f (α2 , β1 ) f (α2 , α2 ) f (α2 , β2 ) ··· 0 0 0 ··· 0 

 f (β2 , α1 ) f (β2 , β1 ) f (β2 , α2 ) f (β2 , β2 ) ··· 0 0 0 ··· 0
 

.. .. .. .. .. .. .. .. ..
 
.
 

 . . . . . . . . ··· 


 0 0 0 0 ··· f (αk , αk ) f (αk , βk ) f (α1 , γ1 ) ··· 0 


 0 0 0 0 ··· f (βk , αk ) f (βk , βk ) f (βk , γ1 ) ··· 0 

0 0 0 0 ··· f (γ1 , αk ) f (γ1 , βk ) f (γ1 , γ1 ) ··· 0
 
 
.. .. .. .. .. .. .. .. ..
 
.
 
 . . . . . . . . 
0 0 0 0 ··· 0 0 0 ··· f (γs , γs )

Which is the announced form. The rank of this matrix is 2k

k is dependent on the exact decomposition from above and the decomposition is not unique.

Remark on the determinant of this matrix

To calculate the determinant we change every second vector so we get a diagonal matrix with
if 2k = n we get det(s) = 1 otherwise there are zeros on the diagonal.

For every skew-symmetric bilinear b form over every basis of V there is a basis change matrix
p such that b = pspT with s in the form from above.

1. if dim(V ) is odd det(s) = 0 so det(b) = 0

2. if dim(V ) is even and det(s) = 1 it follows det(b) = det(p) det(s) det(pT ) = det(p)2

7
 Georg Schöckle

5 Linear-forms
5.1 Theorem
Let f be any bilinear form on any vector space and and any field. Let Lf and Rf be the
independent mappings of f . f is skew-symmetric if and only if Lf = −Rf .
Proof: First let f be skew-symmetric it holds:

Lf (u)(v) = f (u, v) = −f (v, u) = −Rf (u)(v)

Let Lf = −Rf :
f (u, v) = Lf (u)(v) = −Rf (u)(v) = −f (v, u)

5.2 Theorem
Let L1 , L2 be linear forms. Let f (α, β) = L1 (α)L2 (β) − L1 (β)L2 (α). f is a skew-symmetric
bilinear form. Proof:

f (α, β) = L1 (α)L2 (β) − L1 (β)L2 (α) = −L1 (β)L2 (α) + L1 (α)L2 (β) = −f (β, α)

8
 Georg Schöckle

References
[HK71] Kenneth Hoffman and Ray Kunze. Linear Algebra. Prentice-Hall, Englewood
Cliffs, N.J, 1971.
[MH73] John Milnor and Dale Husemoller. Symmetric Biliear Forms. Springer Verlag,
N.Y, Heidelberg, Berlin, 1973.
[Jac09] Nathan Jacobson. Basic Algebra I. Dover Publication, Mineola, N.Y, 2009.