REAL ORTHOGONAL SPACES AND

ORTHOGONAL GROUPS
An Introduction to the Concepts Required in the
Theory of the Representation of the Euclidean
Group.
José Marı́a Rico
Instituto Tecnológico de Celaya,
Departamento de Ingenierı́a Mecánica
Celaya, Gto. 38000, México
May 14, 2018

1 Preface
The purpose of these notes is to gather in a single work concepts and results
that play an important role in the theory of representations of the Euclidean
group, and hence in the kinematics of the rigid body. Most of the results
presented here can be found scattered in the specialized literature. However, as
far as the author is aware, Theorem 7 in Section 4 is a new generalization of a
result by Porteous. Furthermore, most of Section 7, on orthogonal mappings of
degenerate spaces, is thought to be new.

2 Definitions and Invertible Elements.

The concept of a real orthogonal space X is most conveniently introduced in
terms of a symmetric scalar product, also called symmetric bilinear form, on a
real vector space X.
Definition 1. Symmetric Bilinear Form. A symmetric bilinear form on
X is a bilinear map
X × X −→ R (a, b) −→ a · b
such that
(a, b) = a · b = b · a = (a, b) ∀a, b ∈ X.
Since the map is bilinear

(µ1 a1 + µ2 a2 , b) = µ1 (a1 , b) + µ2 (a2 , b),

1
 (a, µ1 b1 + µ2 b2 ) = µ1 (a, b1 ) + µ2 (a, b2 ).
Since the vector space is real, and the mapping is upon the real field, the sym-
metric bilinear form is called real. Only real symmetric bilinear forms will be
considered here.
Definition 2. Cuadratic Form. The quadratic form associated with a
symmetric bilinear form is the map

X −→ R a −→ a · a = (a, a).

This definition implies that a symmetric bilinear form uniquely defines a quadratic
form. The following results simply shows that the opposite is also true.
Theorem 3. A symmetric bilinear form is uniquely defined by its quadratic
form.
Proof. Let (a − b) ∈ X; then

(a − b, a − b) = (a − b, a) − (a − b, b) = (a, a) − (b, a) − (a, b) + (b, b)

= (a, a) − 2(a, b) + (b, b).

Therefore
1
(a, b) =
[(a, a) + (b, b) − (a − b, a − b)]
2
Corollary 4. The symmetric bilinear form is the zero map if, and only if,
its quadratic form is the zero map.
Definition 5. Two elements a, b ∈ X are said to be mutually orthogonal if

(a, b) = a · b = 0.

Two subsets A, B ⊂ X are said to be mutually orthogonal if

(a, b) = a · b = 0 ∀a ∈ A, b ∈ B.

Definition 6. A real vector space X with a real, symmetric bilinear form

will be called a real orthogonal space. Any vector subspace W of an orthogonal
space X is an orthogonal subspace of X under the restriction to W × W of the
symmetric bilinear form of X.
Definition 7. An orthogonal space X is said to be positive-definite (positive-
semidefinite) if, for all non-zero

a ∈ X1 (a, a) > 0 ((a, a) ≥ 0).

Conversely the orthogonal space is said to be negative-definite (negative-semidefinite)

if for all non-zero
a ∈ X (a, a) < 0 ((a, a) ≤ 0).
1 Since the symmetric bilinear form is bilinear, then for
0∈X (0, 0) = (0a, 0b) = 0(a, b) = 0.

2
Finally, an orthogonal space is said to be indefinite if there exists a, b ∈ X such
that
(a, a) > 0 and (b, b) < 0.
An orthogonal space whose symmetric scalar product (symmetric bilinear form)
is the zero map is said to be isotropic, and an orthogonal space that is the direct
sum of two isotropic subspaces is said to be neutral.
Examples.

1. The vector space Rp+q together with the symmetric bilinear form
p
X q
X
(a, b) = − a i bi + ap+j bp+j .
i=1 j=1

has the structure of a real orthogonal space, and it will be denoted by

Rp,q .

2. The vector space R2n together with the symmetric bilinear form
n
X
(a, b) = (ai bn+i + an+i bi )
i=1

has the structure of a real orthogonal space, and it will be denoted by

Rhb2n , the suffix hb stands for hyperbolic, in particular, Rhb2 is the stan-
dard hyperbolic plane.

Definition 8. An a ∈ X is said to be invertible, also called non-null if

(a, a) 6= 0. The element
a
a−1 = .
(a, a)
is called the inverse of a.
Definition 9. Every non-isotropic real orthogonal space has invertible ele-
ments.
Theorem 10. Let a ∈ X be invertible; then, for some µ ∈ X, (µa, µa) = ±1.
Proof. Let a ∈ X be invertible; then

(µa, µa) = µ2 (a, a).

Since a is invertible, then (a, a) 6= 0. Therefore, the equation

1
µ2 = ±
(a, a)

always has a real solution.

Theorem 11. If a and b are invertible elements of an orthogonal space X,
and (a, a) = (b, b), then a + b and a − b are mutually orthogonal, and either a + b
or a − b is invertible.

3
 Proof. Firstly consider

(a + b, a − b) = (a, a − b) + (b, a − b) = (a, a) − (a, b) + (b, a) − (b, b)

= (a, a) − (b, b) = 0

Hence, a + b and a − b are orthogonal. Secondly consider

(a + b, a + b) + (a − b, a − b) = (a, a) + 2(a, b) + (b, b) + (a, a) − 2(a, b) + (b, b)

= 2(a, a) + 2(b, b) = 4(a, a).

Since (a, a) 6= 0, then (a + b, a + b) and (a − b, a − b) cannot be both zero; thus,

at least one of them must be invertible.

3 Linear Correlations and Degenerate Spaces.

Definition 1. Let X be any real vector space, with X L as its dual space; this
is to say X L is the vector space constituted by all the linear mapppings from X
to the real field R; these mappings are also called linear functionals. A linear
correlation on X is any linear map

σ : X −→ X L σ(x) = xσ

The image of x under σ is a linear functional represented by xσ . A correlation

is said to be symmetric if

aσ (b) = bσ (a) ∀a, b ∈ X.

Example. Consider R3 as a vector space, and let x = (x1 , x2 , x3 ), y =

(y1 , y2 , y3 ) ∈ R3 . Define a correlation σ : R3 −→ R3L such that σ(x) = xσ ,
where xσ : R3 −→ R is defined as follows

xσ (y) = 2x1 y1 − 3x2 y3 + 5x3 y2 ,

or
xσ ( ) = 2x1 ( )1 − 3x2 ( )3 + 5x3 ( )2 .
This is a linear correlation. However it is not symmetric.2
Theorem 2. Any linear symmetric correlation σ induces a symmetric bi-
linear form in the following way

(a, b) = aσ (b) = bσ (a).

2 Of course there are non-linear correlations, assume that in the example σ : R3 −→ R3L

is defined such that xσ : R3 −→ R is

xσ (y) = y1 x23 − 3y2 x1 x2 ,
then this correlation is non-linear.

4
Conversely, any scalar product (a, b) induces a unique symmetric correlation,
namely the map a −→ aσ , where the functional aσ is defined by

aσ (b) = (a, b) ∀b ∈ X.

Proof. Assume σ is a linear symmetric correlation, and let a, b ∈ X. Then

define a form as follows
(a, b) = aσ (b).
It suffices to show that this form is symmetric and bilinear, consider

(a, b) = aσ (b) = bσ (a) = (b, a),

the form is symmetric. Let b1 , b2 ∈ X; then3

(a, b1 + µb2 ) = aσ (b1 + µb2 ) = aσ (b1 ) + aσ (µb2 ) = (a, b1 ) + µ(a, b2 ).

Assume that (a, b) is a symmetric bilinear form in X, and define the correlation
σ : X −→ X L as σ(a) = aσ where aσ is defined as

aσ (b) = (a, b).

Since the form is symmetric, then

aσ (b) = (a, b) = (b, a) = bσ (a),

and the correlation is symmetric. Since the form is bilinear, then

aσ (b1 + µb2 ) = (a, b1 + µb2 ) = (a, b1 ) + µ(a, b2 )

= aσ (b1 ) + µ[aσ (b2 )].

Thus the mapping aσ is a linear functional. It remains to show that σ is linear.

Consider

(a1 + µa2 )σ (b) = (a + µa2 , b) = (a1 , b) + µ(a2 , b)

= aσ1 (b) + µaσ2 (b).

This last result completes the proof.

A real linear space X with a correlation σ is called a real correlated vec-
tor space. Since, by the previous theorem, any real orthogonal space has an
induced symmetric correlation, any real orthogonal space can be regarded as
a real correlated vector space, where the correlation is, in addition, linear and
symmetric. The converse result is also true.
The following results require that the vector space involved be finite-dimensional.
This is the class of vector space that we will deal with in our studies. However,
some of the results are also valid for infinite-dimensional vector spaces.
3 Since the form is symmetric, in order to prove its bilinearity it is enough to prove that

the form is linear in one of its two arguments.

5
 Definition 3. Let X be a finite-dimensional real orthogonal space with
induced correlation σ. Since σ : X −→ X L , and dimX = dimX L , then σ is
injective if, and only if, σ is bijective. If σ is bijective, then the orthogonal
space, its symmetric bilinear form, its quadratic form, and its correlation are
said to be non-degenerate or non-singular. If, on the other hand, σ is not
bijective, then σ is not injective, and ker σ 6= {0}. Then X, the correlation σ
and the induced symmetric bilinear form are said to be degenerate, and the
kernel of σ is also called the kernel4 of X.
Theorem 4. The kernel of X is a subspace, and it contains each and every
one of the elements of X which are orthogonal to X itself.
Proof. First, it will be enlightening to characterize explicitly the elements
of the kernel of X. An a ∈ X belongs to the kernel of σ if, and only if, σ(a) = aσ
is the zero functional, this is to say aσ (b) = 0 for every b ∈ X. Consider now
a, b ∈ kerX; then
σ(a + b) = σ(a) + σ(b) = 0 + 0 = 0
Hence a + b ∈ kerX. Similary, if µ ∈ R, then

σ(µa) = µσ(a) = µ0 = 0.

Hence µa ∈ kerX, and kerX is a subspace of X. Next, let a ∈ kerX; then

σ(a) = aσ = 0, therefore

aσ (x) = (a, x) = 0 ∀x ∈ X;

hence a is orthogonal to X. Finally, assume that a ∈ X satisfies that (a, x) =

0 ∀x ∈ X; then
aσ (x) = 0 ∀x ∈ X,
hence aσ is the zero functional, and a ∈ kerX.
Example. Consider R3 and x = (x1 , x2 , x3 ), y = (x1 , x2 , x3 ) ∈ R3 define a
correlation σ : R3 −→ R3L such that σ(x) = xσ , and

xσ (y) = 2y1 x1 − 3y3 x3

or
xσ ( ) = 2x1 ( )1 − 3x3 ( )3
3
Then x ∈ kerℜ if, and only if, x is the zero map; i.e. xσ (y) = 0 for every
σ

y ∈ ℜ3 ; since e1 , e2 , e3 is a basis of ℜ3 , then x ∈ ℜ3 if, and only if,

xσ (ei ) = 0 f or i = 1, 2, 3.

Therefore

xσ (e1 ) = 2(1)x1 − 3(0)x3 = 2x1 = 0

σ
x (e2 ) = 2(0)x1 − 3(0)x3 = 0
xσ (e3 ) = 2(0)x1 − 3(1)x3 = −3x3 = 0
4 The kernel is also called the radical.

6
Thus x ∈ kerℜ3 if, and only if, x1 = x3 = 0. Hence

kerℜ3 = {x | x = (0, x2 , 0) for x2 ∈ ℜ} = [(0, 1, 0)].

Corollary 5. A positive-definite (negative-definite) orthogonal space is non-

degenerate.
Proof. Let X be a positive-definite orthogonal space; then for every non-
zero a ∈ X (a, a) > 0 hence (a, a) 6= 0. Therefore aσ (a) 6= 0, and aσ is not the
zero functional. Thus a 6∈ kerX, and σ is inyective and biyective. The proof for
negative-definite orthogonal spaces follows the same argument.
Theorem 6. Let A be a finite set of mutually orhogonal invertible elements
of a real orthogonal space X. Then the vector subspace generated by A, also
called its linear span, is a non-degenerate orthogonal subspace of X. Here, it
will be denoted by R[A].
Proof. Let x ∈ R[A] be non-zero; then x = Σµi ai , where ai ∈ A, and at
least one µi is different from zero. Since the ai ’s are invertible, it is possible to
define
y = Σµi a−1
i

Consider then

(x, y) = (Σµi ai , Σµi a−1

i ) = (Σµi ai , Σµi (ai , ai )
−1
ai )

Since the a′i s are orthogonal (ai , aj ) = 0 for i 6= j. Then

(x, y) = Σµ2i (ai , ai )−1 (ai , ai ) = Σµ2i 6= 0.

Thus x 6∈ kerR[A] and R[A] is non-degenerate.

Corollary 7. For any finite p, q the orthogonal space Rp,q is non-degenarate.
Proof. Consider the trivial basis {e1 , . . . , ep , ep+a , . . . , ep+q }; then

(ei , ei ) = −1 for every i = 1, ..., p

and
(ei , ei ) = +1 for every 1 = p + a, ..., p + q
Hence the e′i sare invertible and the conditions of the previous theorem apply.
Theorem 8. Let X be a finite-dimensional real orthogonal space, and let
X ′ be a linear complement in X of kerX. Then X ′ is a non-degenarate subspace
of X.
Proof. Assume by the way of contradiction, that there exists a non-zero
x ∈ X ′ such that
(x, y) = 0 ∀y ∈ X ′
Since, by definition of kerX, (x, z) = 0 for every z ∈ kerX, and X = kerX ⊕ X ′
(this assumption implies that X ′ ∩ kerX = {0} then

(x, w) = 0 ∀w ∈ X.

This result is a contradiction to x ∈ X ′ .

7
4 Orthogonal Maps.
Definition 1. Let X an Y be real orthogonal spaces with induced correlations
σ and τ respectively. A map t : X → Y is said to be an orthogonal map if it is
linear and, for all a, b ∈ X

[t(a)]τ [t(b)] = aσ (b).

This condition can be expressed in terms of the corresponding symmetric bilin-

ear forms as
(t(a), t(b))Y = (a, b)X
where the suffixes Y and X indicate the symmetric bilinear form involved in
each computation.
Theorem 2. If X is non-degenarate, then any orthogonal map t : X → Y
is injective.
Proof. Assume, by the way of contradiction, that there exist a, b ∈ X such
that a 6= b, but t(a) = t(b). Then a − b 6= 0, but t(a − b) = t(a) − t(b) = 0; thus

(t(a − b), y) = (0, y) = 0 ∀y ∈ Y.

Since t : X → Y , then for every x ∈ X, t(x) ∈ Y . Thus for every x ∈ X

0 = (t(a − b), y) = (t(a − b), t(x)) = (a − b, x)

Thus a−b ∈ KerX; since a−b 6= 0, this is a contradiction to the non-degeneracy

of X.
The following example shows that the non-degeneracy condition is necessary.
Example 3. Consider the vector space R3 together with the symmetric
bilinear form

(a, b) = a1 b1 + a2 b2 , where a = (a1 , a2 , a3 ), b = (b1 , b2 , b3 ) ∈ R3 .

Then consider the mapping

t : R3 → R3 , defined by t(a1 , a2 , a3 ) = (a1 , a2 , 0).

A simple calculation shows that t is not injective since

t(0, 0, a3 ) = (0, 0, 0) ∀a2 6= 0.

However

(t(a), y(b)) = (t(a1 , a2 , a3 ), (t(b1 , b2 , b3 ) = ((a1 , a2 , 0), (b1 , b2 , 0))

= a1 b1 + a2 b2 = (a, b).

Theorem 4. Let W, X, and Y be orthogonal spaces, and let t : X → Y

and u : W → X be orthogonal maps. Then IX is orthogonal, tu : W → Y is
orthogonal, and if t is invertible, t−1 is orthogonal.

8
 Proof. The linearity of these mappings is easily proved. Here, only the
preservation of the symmetric bilinear form will be proved. Let a, b ∈ X be
arbitrary; the IX (a) = a, and IX (b) = b thus

(IX (a), IX (b)) = (a, b).

Let x, y ∈ W be arbitrary; then

(tu(x), tu(y)) = (u(x), u(y)) = (x, y).

Assume t : X → Y is invertible; then there exists a linear mapping t−1 : Y → X,

such that t−1 t = tt−1 = IX . Let y1 , y2 ∈ Y and x1 , x2 ∈ X be elements such
that t(x1 ) = y1 , and t(x2 ) = y2 ; then

(x1 , x2 ) = (t(x1 ), t(x2 )) = (y1 , y2 )

Thus
(y1 , y2 ) = (x1 , x2 ) = (t−1 (y1 ), t−1 y2 ),
and t−1 is orthogonal.
Definition 5. An invertible orthogonal map t : X → Y will be called an
orthogonal isomorphism, and the two orthogonal space X and Y are said to be
isomorphic.
2n
Theorem 6. For any finite n, the orthogonal spaces Rn,n and Rhb are
isomorphic.
Definition 6. An invertible orthogonal map t : X → X is called an or-
thogonal automorphism of the orthogonal space. It is important to recall
that since both, the range and the image of the mapping t are the same, the
invertibility of the mapping is assured by the injectivity or the surjectivity of
the mapping.
Theorem 8. Any orthogonal mapping of a non-degenerate finite-dimensional
orthogonal space X into itself is an orthogonal automorphism of X.
Proof. By Theorem 2, any orthogonal mapping of a non-degenerate finite-
dimensional orthogonal space X is injective; therefore the mapping is surjective
and bijective.
Definition 9. Let X and Y be orthogonal spaces; the set of orthogonal
maps t : X → Y will be denoted by O(X, Y ). The set of the orthogonal
automorphisms t : X → X will be denoted by O(X).
Theorem 10. O(X) forms a group under the operation of composition.
This group is called the group of orthogonal automorphisms of X.
Proof. It is well known that mappings, in general, form a group under the
operation of composition. Therefore, only the closure and the presence of the
indentity will be indicated.

• Let s, t ∈ O(X), then st : X → X and according to Theorem 4, st is also

orthogonal. It remains to show that st is invertible. Consider t−1 s−1 ;
then
(st)(t−1 s−1 ) = s(tt−1 )s−1 = sIX s−1 = IX

9
 and
(t−1 s−1 )st = t−1 (s−1 s)t = t−1 IX t = IX
Thus st ∈ O(X).

• It is easily shown that IX is an invertible orthogonal map of X, thus

IX ∈ O(X).

• Finally, if s ∈ O(X), then s is invertible, and there exists a mapping

s−1 : X → X; according to theorem 4, s−1 is orthogonal, and evidently is
invertible for
ss−1 = 1X = s−1 s

The groups of orthogonal automorphisms of special orthogonal spaces have

particular denominations. For example

O(Rp,q ) = O(p, q; R) ≡ O(p, q),

O(R0,n ) ≡ (n; R) ≡ O(n).

Definition 11. Let X be an arbitrary vector space; a linear transformation
t : X → X is said that preserves the orientation of the vector space if det(t) > 0;
on the contrary, it is said that reverses the orientation of the vector space if
det(t) < 0. It is important to notice that his definition implicity requires that t
be invertible.
Definition 12. The orientation preserving orthogonal automorphisms are
called special orthogonal automorphisms or rotations of X. The orien-
tation reversing orthogonal automorphisms of X are called antirotations or
reflections of X.
Theorem 13. The set of special orthogonal automorphisms of X forms a
normal subgroup of the group of orthogonal automorphisms.
Definition 14. the subgroup of orientation preserving orthogonal automor-
phisms of th orthogonal space X is denoted by SO(X). In particular

SO(Rp,q ) ≡ SO(p, q; R) ≡ SO(p, q),

SO(R0,n ) ≡ SO(n; R) ≡ SO(n).

5 Orthogonal Annihilators
Definition 1. Let W be a subspace of a vector space X, the dual annihilator
of W , denoted W @ , is the subspace of the dual subspace X L that annihilates
W ; i.e.
W @ = {β ∈ X L | β(w) = 0 ∀w ∈ W }
Theorem 2. Let W be a subspace of a vector space X; then

dimW @ = dimX − dimW.

10
 Proof. First it is necessary to recall that dimX = dimX L , and arbitrary
element of the dual space is of the form

f (x) = a1 x1 + a2 x2 + . . . + an xn with dimX = n,

where x = (x1 , x2 , . . . , xn ) ∈ X, and the constants a1 , a2 , . . . , an determine the

functional. Let {w1 , w2 , . . . , wk } be a basis of W , then f (W ) = 0 if, and only
if, 5
f (wi ) = 0 for i = 1, 2, . . . , k.
The equation above represents k equations in n unknowns. Therefore the
solution space, and thus the dual annihilator of W has dimension n − k =
dimX − dimW .
Definition 3. Let W be an orthogonal subspace of an orthogonal space X,
and let σ the induced linear correlation, then the orthogonal annihilator of
W , denoted by W ⊥ , is defined as

W ⊥ = σ ⊣ (W @ ) = {x ∈ X | σ(X) ∈ W @ }

Remark. Since σ is a linear mapping then it can be proved that W ⊥ is also a

subspace. Furthermore dimW ⊥ ≥ dimW @ = dimX − dimW , and dimW ⊥ =
dimW @ = dimX − dimW only when σ is an injection and therefore a bijection.
The following result provides a simpler characterization of the orthogonal
annihilator of a subspace W of an orthogonal space X.
Theorem 4. (Characterization of the Orthogonal Annihilators) The or-
thogonal annihilator contains each and every one of the elements of X that are
orthogonal to the complete subspace W .
Proof. Assume that a ∈ X is orthogonal to W , and σ is the induced linear
correlation, then
aσ (b) = (a, b) = 0 ∀b ∈ W,
Thus aσ ∈ W @ . Furthermore, let aσ ∈ W @ then

aσ (b) = (a, b) = 0 ∀b ∈ W

Thus a ∈ W ⊥ .
Definition 5. Let X be a real orhogonal space and W a subspace of X with
linear complement Y ; i.e. X = W ⊕ Y ; if Y is contained in W ⊥ then we said
that Y is an orthogonal complement of W in X. A direct sum decomposition
X = W ⊕ Y is said to be an orthogonal decomposition.
Example. Consider the orthogonal space R4 with the symmetric bilinear
form
(x, y) = 0x1 y1 + x2 y2 + x3 y3 + x4 y4
Consider W = [(1, 0, 0, 0)]. Then, the orthogonal annihilator of W is W ⊥ = R4 ,
and Y = {(0, y2 , y3 , y4 ) | y2 , y3 , y4 ∈ ℜ} is an orthogonal complement of W in
R4 .
5 Furthermore it can be proved that the equations are independent.

11
 Theorem 6. Let W be a linear subspace of a finite-dimensional real or-
thogonal space X. Then kerW = W ∩ W ⊥ .
Proof. By definition

kerW = {w | w ∈ W and (w, w′ ) = 0 ∀w′ ∈ W };

then
kerW = {w | w ∈ W } ∩ {w | (w, w′ ) = 0 ∀w′ ∈ W }.
Thus
kerW = W ∩ W ⊥ .
Theorem 7. Let W be a linear subspace of a finite-dimensional real or-
thogonal space X. Then X = W ⊕ W ⊥ if, and only if, W is non-degenerate; in
this case, W ⊥ is the unique orthogonal complement of W in X.
Proof. Assume W is degenerate then W ∩ W ⊥ 6= {0}, and the sum is not
direct. Assume W is non-degenerate. Then W ∩ W ⊥ = {0}; thus the sum
W + W ⊥ es direct. Then

dimW @ = dimX − dimW ;

since
dimW ⊥ = dim[σ ⊣ (W @ )] ≥ dimW @ .
where σ is the induced correlation. Thus

dimW ⊥ ≥ dimX − dimW.

Then

dim(W +W ⊥ ) = dimW +dimW ⊥ −dim(W ∩W ⊥ ) = dimW +dimW ⊥ ≥ dimX;

Furthermore, W + W ⊥ is a subspace of X, and thus

dim(W + W ⊥ ) ≤ dimX.

Therefore
W + W ⊥ = X.
The uniqueness of the orhogonal complement is proved by contradiction. As-
sume that U1 , and U2 are two orthogonal complements, then

X = W ⊕ U1 with U1 ⊂ W ⊥

and
X = W ⊕ U2 with U2 ⊂ W ⊥ .
Let {w1 , w2 , . . . , wk }, {u1 , u2 , . . . , un−k }, and {u′1 , u′2 , . . . , u′n−k } be bases of
W, U1 , and U2 respectively. Since, by assumption U1 6= U2 , there is at least

12
one u′i , let us say u′1 , such that u′1 6∈ U1 . Since X = W + U1 , then there exist
w ∈ W and u ∈ U1 , such that

u′1 = w + u then u′1 − u = w.

Since u′1 ∈ U2 , and u ∈ U1 , and both subspaces are contained in W ⊥ , then

(u′1 −u) ∈ W ⊥ ; furthermore (u′1 −u) ∈ W ∩W ⊥ . However W is non-degenerate,
thus W ∩ W ⊥ = {0}, and u′1 = u contradicting the assumption.
Corollary 8. Let a be an element of a real orthogonal space X. Then
X = (R[a]) ⊕ (R[a])⊥ if, and only if, a is invertible.
Proof. By the previous theorem, the statement is true if, and only if, R[a] is
a non-degenerate subspace of X. Assume that a is invertible, then an arbitrary
non-zero element of R[a] is given by µa, and

(µa, µa) = µ2 (a, a) 6= 0.

Thus R[a] is non-degenerate.

Assuming that a is not invertible; then two arbitrary elements of R[a] are
given by µa, and δa for µ, δ ∈ R. Thus

(µa, δa) = µδ(a, a) = µδ(0) = 0

and R[a] is degenerate.

Theorem 9. Let W be a subspace of a non-degenerate finite-dimensional
real orthogonal space X. Then (W ⊥ )⊥ = W .
Proof. Let w ∈ W and w′ ∈ W ⊥ be arbitrary, then

(w, w′ ) = 0 ∀w′ ∈ W ⊥ .

Thus w ∈ (W ⊥ )⊥ , and W is contained in (W ⊥ )⊥ . Furthermore, since X is non-

degenerate, then

dim(W ⊥ )⊥ = dimX − dimW ⊥ = dimX − (dimX − dimW ) = dimW

Thus (W ⊥ )⊥ .
Corollary 10. Let W be a non-degenerate orthogonal subspace of a non-
degenerate orthogonal space X, then W ⊥ is also a non-degenerate orthogonal
subspace of X.
Proof. Since W is non-degenerate, then

W ∩ W ⊥ = {0},

and by theorem 9,
(W ⊥ )⊥ = W.
Thus
W ⊥ ∩ (W ⊥ )⊥ = {0},
and W ⊥ is non-degenerate.

13
6 Basis of Orthogonal Spaces.
Theorem 1. Basis Theorem for Real Orthogonal Spaces. An non-
dimensional non-degenerate real orthogonal space X, with n > 0, is express-
ible as the direct sum of n non-degenerate mutually orthogonal lines (one-
dimensional subspaces of X).
Proof. By induction, if n = 1, the space consists of only one line, and by
assumption the line is non-degenerate. Assume that the theorem is true for n
and consider the case when the dimension is n + 1.
Since the space is non-degenerate, by the theorem 2.9, there exists and in-
vertible element a ∈ X that defines a non-degenerate line R[a], such that
X = (R[a]) ⊕ (R[a])⊥
By the corollary 5.10 (R[a])⊥ is also non-degenarate and of dimension n, thus
the proof is finished.
Definition 2. A linearly independent subset S of a real orthogonal space
X is said to be an Orthonormal Subset of X if two distinct elements of S
are mutually orthogonal, and the quadratic form of any element of S is equal
to −1, 0, or 1. If S also spans X, then S is said to be an Orthonormal Basis
of X.
Theorem 3. Basis Theorem. Any finite-dimensional orthogonal space X
has an orthonormal basis.
Proof. Let X ′ be a linear complement, in X, of kerX. Then X ′ is a non-
degenarate subspace of X, and, by theorem 1, it has a basis B ′ such that is
orthogonal and for any b′i ∈ B ′ satisfies (b′i , b′i ) 6= 0. Then, by applying theorem
2.10 it is possible to find a basis B, of X ′ , orthogonal and with
(bi , bi ) ∈ (−1, 1)
Let A be any basis for kerX, then A ∪ B is an orthogonal basis, such that their
elements ci satisfy
(ci , ci ) ∈ (−1, 0, 1)
Thus A ∪ B is an orthonormal basis of X.
Corollary 4. Any non-degenerate finite-dimensional orthogonal space X is
isomorphic to Rp,q for some finite p, and q.
Corollary 5. Any orthogonal finite-dimensional space X is isomorphic to
Rp,q,r , where the symmetric bilinear form is
(Rp,q,r , Rp,q,r ) = R
p
X q
X r
X
(x, y) = x i yi − xp+j yp+j + xp+q+k yp+q+k
i=1 j=1 k=1
Furthermore, the orthogonal space X can be orthogonally decomposed as
X = kerX ⊕ W1 ⊕ W2
where dim(kerX) = p, dimW1 = q, and dimW2 = r. Further W1 is negative
definite, and W2 is positive definite.

14
7 Reflections and Orthogonal Mappings
Theorem 1. Let W ⊕ Y be an orthogonal decomposition of an orthogonal
space X. Then the mapping

σ : X → X; x=w+y w−y

where x ∈ W , amd y ∈ Y , is linear and orthogonal.

Proof. Let v1 , v2 ∈ X be arbitrary, then there exist w1 , w2 ∈ W , and
y1 , y2 ∈ Y such that

v 1 = w1 + y1 , and v 2 = w2 + y2 .

Then σ(v1 ) = w1 − y1 , and v2 = w2 + y2 . Thus

(σ(v1 ), σ(v2 )) = (w1 − y1 , w2 − y2 ) = (w1 , w2 ) − (w1 , y2 ) − (y1 , w2 ) + (y1 , y2 )

= (w1 , w2 ) + (y1 , y2 ) = (v1 , v2 ).

Definition 2. The map analized in the previous theorem is called a reflection

of X in W . When Y = W ⊥ , the map is said to be bf the reflection of X in
W . A reflection of X in a linear hyperplane W is said to be a ryperplane
reflection of X.
Theorem 3. A hyperplane reflection is a bijection. Furthernore, a hy-
perplane reflection is an involution; i.e. the reflection is its own inverse. In
particular these results apply to hyperplane reflections.
Theorem 4. A hyperplane reflection of a finite-dimensional orthogonal
space X is an antirotation of X.
Proof. Let W be a hyperplane of X, and let {w1 , w2 , . . . , wn−1 } be a basis
for W . Then there exists a xn ∈ X such that {w1 , w2 , ..., wn−1 } ∪ {xn } is a basis
of X. The image of the elements of this basis, under the hyperplane reflection
rw are
rw (wi ) = wi ∀ i = 1, 2, ..., n − 1
rw (xn ) = −xn
Therefore, the matrix representative of the mapping is
 
1

 1 0 

 . 
Rw =  

 . 

 0 1 
−1

Thus (detRw ) = −1, and rw is an antirotation.

Remark. If a is an invertible element of a finite-dimensional orthogonal space
X. Then the orthogonal complement of R[a] is unique and denoted by (R[a])⊥ .
The reflection of X in the hyperplane (R[a])⊥ will be denoted by pa .

15
 Theorem 5. Suppose that a and b are invertible elements of a finite-
dimensional orthogonal space X, such that (a, a) = (b, b). Thena may be
mapped into b either by a single hyperplane reflections of X.
Proof. By theorem 2.11, either a − b or a + b are invertible. Furthermore
a − b and a + b are orthogonal since

(a − b, a + b) = (a, a) + (a, b) − (b, a) − (b, b) = (a, a) − (b, b) = 0

Assuming that a − b is invertible, consider

1 1 1 1
pa − b(a) = pa − b[ (a − b) + (a + b)] = − (a − b) + (a + b) = b
2 2 2 2
If a − b is not invertible, then a + b must be invertible. Then consider
1 1 1 1
pb pa + b(a) = pb pa + b[ (a − b) + (a + b)] = pb [ (a − b) − (a + b)]
2 2 2 2
Theorem 6. Any orthogonal transformation t : X −→ X of a non-
degenerate finite-dimensional space X is expressible as the composite of a finite
number of hyperplane reflections of X; the number of reflections is not greater
than 2dimX, or if X is positive-definite (negative-definite) dimX.
Proof. The theorem is obvious for dimX = 1. Suppose, by the hypothesis
induction, that the theorem is true for dimX = n, and consider the case dimX =
n + 1.
Let {e1 , e2 , . . . , en+1 } be an orthonormal basis of X, and let t : X −→ X
be any orthogonal mapping of X. Then (ten+1 , ten+1 ) = (en+1 , en+1 ); thus,
by theorem 5, there exists an orthogonal mapping u : X −→ X, formed by
either a hyperplane reflection or the composite of two hyperplane reflections
of X (Furthermore if the space is definite, then the orthogonal mapping can
always be expressed by a single hyperplane reflection) such that ut(en+1 ) =
en+1 . Consider now the mapping ut : X −→ X, since it is the composition of
orthogonal mappings, then it is also orthogonal. Furthermore ut(en+1 ) = en+1 ,
and since X = [e1 , e2 , ..., en ] ⊕ [en+1 ], then the restriction of ut to [e1 , e2 , ..., en ]
maps the subspace into itself; i.e.

(ut)R : [e1 , e2 , . . . , en ] −→ [e1 , e2 , . . . , en ].

Thus (ut)R is an orthogonal map of an n-th dimensional orthogonal space, and

the induction hypothesis applies. Finally, the mapping t can be written as

t = u−1 (ut)R ,

and the result follows.

Corollary 7. An orthogonal mapping t of a non-degenerate orthogonal
space X satisfies (det t) = ±1. Furthermore, if t is a rotation (det t) = 1. In
other words, if X is a non-degenerate orthogonal space, then t ∈ O(X) implies
that (det t) = ±1, and t ∈ SO(X) implies (det t) = +1.

16
8 Orthogonal Mappings of Degenerate Spaces
Corollary 5, section 6, establishes that any finite-dimensional orthogonal space
X can be orthogonally decomposed as

X = kerX ⊕ W1 ⊕ W2

The first step in the analysis of orthogonal mappings of degenerate spaces consist
in analyzing the effect of the mapping into these subspaces.
Theorem 1. Let t : X −→ X be an orthogonal mapping of a degener-
ate orthogonal space, then kerX, W1 , and W2 are invariant subspaces of t.
Furthermore, t(W1 ) = W1 , and t(W2 ) = W2 .
Proof. Let {e1 , ..., ep , ep+1 , . . . , ep+q , ep+q+1 , . . . , en } be an orthonormal ba-
sis of X, with the first step p elements as a basis of kerX, the next q elements
as a basis of W1 , and the remaining ones as a basis of W2 . Let x ∈ W1 ; then
q
X
x= µj ep+j
j=1

Assume µ1 t(ep+1 ) + . . . + µq t(ep+q ) = 0, then

t[µ1 ep+1 + . . . + µq ep + q] = 0

Hence,

0 = (0, 0) = (t[µ1 ep+1 + . . . + µq ep+q ], t[µ1 ep+1 + . . . + µq ep+q ])

X
= (µ1 ep+1 + . . . + µq ep+q , µ1 ep+1 + . . . + µq ep+q ) = µ2j (−1)

Thus µ1 = . . . = µq = 0. Therefore the vectors t(ep+1 ), ..., t(ep+q ) are

linearly independent, and thus dim t(W1 ) = dimW1 . Assume that t(W1 ) 6=
W1 , then t(W1 ) ∩ W2 6= {0} or t(W1 ) ∩ kerX 6= {0}; assume the first possibility;
then there exists a w ∈ W1 , such that t(w) ∈ W2 . Then

0 < (w, w) = (tw, tw) < 0

a contradiction. The proof for W2 follows the same lines.

Let x ∈ kerX, and assume by contradiction that t(x) 6∈ kerX. Then

t(x) = y1 + y2 + y3

where y1 ∈ kerX, y2 ∈ W1 , and y3 ∈ W2 . since t(W1 ) = W1 , and t(W2 ) = W2 ,

there is a x2 ∈ W2 such that t(x2 ) = Y2 . Hence

0 = (x, x2 ) = (t(x), t(x2 )) = (y1 + y2 + y3 , y2 ) = (y2 , y2 ) 6= 0

a contradiction, hence t(x) ∈ kerX, and t(kerX) is contained in kerX.

Corollary 3. A mapping t : X → X of a degenerate orthogonal space is an
orthogonal mapping if, and only if, the restriction of t to W = W1 ⊕ W2 is an
orthogonal map, and the restriction of t to kerX is any linear map.

17
 Proof. Assume that t : X → X is an orthofonal map of the degenerate
orthogonal space. Since by theorem 1, W1 and W2 are invariant subspaces,
then W = W1 ⊕ W2 is also invariant, and it is possible to define the restriction
mapping tW : W → W . Furthermore, for every x, y ∈ W

(tW (x), tW (y)) = (t(x), t(y)) = (x, y)

Thus, the restriction is an orthogonal map.

Assume tW : W → W is orthogonal, and tkerX : kerX → X is any linear
map. Let x, y ∈ X be arbitrary, then

(x, y) = (x1 + x2 , y1 + y2 ) = (x1 , y1 ) + (x1 , y2 ) + (x2 , y2 ) = (x2 , y2 )

Furthermore

(t(x), t(y)) = (t(x1 + x2 ), t(y2 + y1 )) = (t(x1 ) + t(x2 ), t(y1 ) + t(y2 ))

= (tkerx (x1 ) + tW (x2 ), tkerX (y1 ) + tW (y2 ))
= (tkerx (x1 ), (tkerx (y1 ) + (tkerx (x1 ), tkerX (y2 ))
+(tW (x2 ), tkerx (y1 ) + (tW (x2 ), tW (y2 ))
= (tW (x2 ), tW (y2 )) = (x2 , y2 ).

Thus equating both equations

(t(x), t(y)) = (x, y),

and the mapping is orthogonal.

Theorem 4. An orthogonal mapping t : X → X of a degenerate orthogonal
space X is bijective if, and only if, the restriction of the mapping t to kerX is
bijective.
Proof. Assume that t : X → X is bijective, then t is injective and kert =
(0), thus ker tkerX = (kert) ∩ kerX = (0). Thus tkerX is injective, and
therefore bijective.
Assume that tkerX is bijective, then

t(kerX) = tkerX (kerX) = kerX.

Furthermore t(W1 ) = W1 , and t(W2 ) = W2 . Thus

t(x) = t(kerX + W1 + W2 ) = kerX + W1 + W2 = X

therefore X is surjective, and hence bijective.

The fact that the restriction of an orthogonal map of a degenerate sapce X
to the kernel of X can be any linear map implies that the results indicated by the
corollary 8 of the section 7 no longer apply. In other words t ∈ O(X) does not
impply that (det t) = ±1, and t ∈ SO(X) does not imply that (det t) = +1.

18
9 Semiorientations and Orthogonal Groups that
Preserve the Semiorientations
Definition 1. Let X be a non-degenerate space; then X is isomorphic to
Rp,q for some integers p, and q. Furthermore, X = W1 ⊕ W2 , where W1 is
negative definite and W2 is positive definite. The semiorientations of X are the
orientations of the subspaces W1 and W2 .
In general, although the elements of SO(X) preserve the orientation of X,
they do not necesarily preserve the orientation of the subspaces W1 and W2
defined in the statement above. The situation is illustrated below.
Example. Consider The orthogonal space R1,2 wich has the symmetric bi-
linear form

(x, y) = ((x1 , x2 , x3 ), (y1 , y2 , y3 )) = −x1 y1 + x2 y2 + x3 y3

Consider now the mapping t : R1,2 → R1,2 given by

t(e1 ) = −e1 ; t(e2 ) = e2 ; t(e3 ) = −e3 ;

It is easy to show that t is orthogonal, since

(t(x), t(y)) = (t(x1 e1 + x2 e2 + x3 e3 ), t(y1 e1 + y2 e2 + y3 e3 ))

= (−x1 e1 + x2 e2 − x3 e3 , −y1 e1 + y2 e2 − y3 e3 )
= x1 y1 (−1) + x2 y2 (+1) + x3 y3 (−1) = −x1 y1 + x2 y2 + x3 y3 = (x, y)

Furthermore, the determinant of the mapping is

 
−1 0 0
| t |= det t =  0 1 0  = 1.
0 0 −1

Thus (det t) = 1, and t ∈ SO(1, 2). However

R1,2 = [e1 ] ⊕ [e2 , e3 ]

where [e1 ] is negative definite, and [e2 , e3 ] is negative-definite. Furthermore t

reverses the orientation of both subspaces [e1 ] and [e2 , e3 ].
Theorem 2. An orthogonal mapping t of a non-degenerate orthogonal
space X = W1 ⊕ W2 , preserves the semiorientation of the space; i.e preserves
the orientations of W1 and W2 . Further , assume that the restrictions of t to
W1 and W2 are invariant subsapces of X under t, then

tW 1 (x) = t(x) ∀ x ∈ W1

and
tW 2 (x) = t(x) ∀ x ∈ W2
then t preserves the semiorientations of W .

19
 Theorem 3. The set of orthogonal automorphisms that preserves the
semiorientatons of a non-degenerate orthogonal space X is a subgroup of SO(X).
This subgroup is denoted by SO+ (X). In particular if X = Rp,q , then SO+ (X)
is denoted by SO+ (p, q). 6

10 Additional Notes to Real Orthogonal Spaces

and Orthogonal Groups
Objective: The notes Real Orthogonal Spaces and Orthogonal Groups, from
now on referred to as the main notes, deal with symmetric bilinear forms without
considering their matrix represenation. These notes attempt to follow a parallel
path through the use of the matrix representation. Further, these notes intent
to show how the matrix representation approach can be reconciled with the
cordinate free approach of the main notes. Since most of the definition can be
found in the main notes, only those strictly indispensable definitions will be
recalled. It is hoped that, by reading these notes, the reader can relate both
approaches and clearly understand any work developed following either of both
approaches.

10.1 Symmetric Bilinear Form and the Matrices Repre-

senting the Form
Proposition 1. Consider an orthogonal space X with a real symmetric form
X × X −→ R. Given a basis B = {x1 , x2 , . . . , xn } of X, there exists a unique
real n-th order symmetric matrix S that represents the symmetric bilinear form
with respect to the basis B. Further, given a n-th order real symmetric matrix
S, and a basis B of X, there exists a unique real symmetric form over X.
Proof. Define the matrix S as follows

sij = (xi , xj ) i, j = 1, 2, ..., n (1)

Then S is an n order real matrix. Further since the form is symmetric

sij = (xi , xj ) = (xj , xi ) = sij i, j = 1, 2, ..., n. (2)

the matrix is symmetric. The other direction of th proof is easy but laborious.
The next proposition shows how the matrix representative of the symmetric
bilinear form can be used to find the value, of the form, for arbitrary vectors of
the space.
Proposition 2. Let X be an orthogonal space with S as the matrix
representative of the symmetric bilinear form with respect to a basis B =
6 The concept of semiorientations can be applied to degenerate orthogonal spaces; however

it is not necessary to consider this possibility.

20
{x1 , x2 , . . . , xn }. Let A = [a1 , a2 , . . . , an ], and C = [c1 , c2 , . . . , cn ] be the co-
ordinate vectors, with respect to the basis B, of two vectors, a, c ∈ X. Then
 
c1
   c2 
 
T 
 . 
(a, c) = [a1 , a2 , . . . , an ] S  
 . 
 (3)
 
 . 
cn

Proof. Since S is the matrix representative of the symmetric bilinear form

with respect to the basis B, then

sij = (xi , xj ) ∀ i, j = 1, 2, . . . , n.

Since

a = a1 x 1 + a2 x 2 + . . . + an x n and c = c1 x1 + c2 x2 + . . . + cn xn ,

then

(a, c) = (a1 x1 + a2 x2 + . . . + an xn , c1 x1 + c2 x2 + . . . + cn xn )
= a1 [c1 (x1 , x1 ) + c2 (x1 , x2 ) + . . . + cn (x1 , xn )] +
a2 [c1 (x2 , x1 ) + c2 (x2 , x2 ) + . . . + cn (x2 , xn )] + . . . +
an [c1 (xn , x1 ) + c2 (xn , x2 ) + . . . + cn (xn , xn )]
= a1 (c1 s11 + c2 s12 + . . . + cn s1n ) +
a2 (c1 s21 + c2 s22 + . . . + cn s2n ) + . . . +
an (c1 sn1 + c2 sn2 + . . . + cn snn )

or
  
s11 s12 . . . s1n c1

 s21 s22 . . . s2n 
 c2 

. . . . . . .
(a, c) = [a1 , a2 , ..., an ]T   = AT SC
  


 . . . . . . 
 . 

 . . . . . .  . 
sn1 sn2 . . . snn cn

One important feature of the use of the matrix representative of the symmetric
bilinear form is that allows one to work with more concrete beings than map-
pings; i.e. matrices and coordinate vectors. However, one drawback is that there
is a multitude of matrix representatives, indeed one for each posible basis of the
orthogonal space. Thus a property of the symmetric bilinear form will be trans-
lated into a common property of all the possible representations. Conversely,
only those properties that are common to all possible matrix representations
are properties of the form.

21
 The next proposition shows the relationship between the different matrix
representatives of the same symmetric bilinear form.
Proposition 3. Let X be an orthogonal space, and let B = {x1 , x2 , . . . , xn },
′ ′ ′ ′
and B = {x1 , x2 , . . . , xn } be two bases of the space. If S is the matrix repre-
sentative of the symmetric bilinear form with respect to the basis B, then the
matrix representative of the symmetric bilinear form with respect to the basis
′ ′
B , denoted by S , is given by
′
S = P SP T ,
′
where P is the transition matrix between the bases B and B .
′
Proof. Since P is the transition matrix from B to B , then
′
x1 = p11 x1 + p12 x2 + . . . + p1n xn
′
x2 = p21 x1 + p22 x2 + . . . + p2n xn
.
.
.
′
xn = pn1 x1 + pn2 x2 + . . . + pnn xn

′ ′
An arbitrary element sij of S is given by
n n n n
′ ′ ′ X X X X
sij = (xi , xj ) = ( pir xr , pjt xt ) = pir (xr , pjt xt )
r=1 t=1 r=1 t=1
n n n n n
" n
#
X X X X X X
= pir pjt (xr , xt ) = pir pjt srt = pir srt pjt
r=1 t=1 r=1 t=1 r=1 t=1
Pn
Notice that t=1 srt pjt is just the (r − th, j − th) element of the product of
matrices M = SP T , denoted mrj . Then,
n
′ X
sij = pir mrj
r=1
′
Again, sij is the (i − th, j − th) element of the product of the matrices P and
M . Hence ′
S = P M = P SP T .
This result provides the foundation for finding what characteristics are common
to all the representations of a symmetric quadratic form, and therefore these
characteristics are properties of the symmetric bilinear form. An important
example of this class of results is now developed.
′
Proposition 4. Let S ans S be two matrices representative pf a symmetric
′
bilinear form, then S is non-singular (singular) if, and only if, S is non-singular
(singular).

22
 Proof. By proposition 3, there exists a non-singular transition matrix P
such that ′
S = P SP T .
Then ′
| S |=| P SP T |=| P || S || P T |
But | P |=| P T |, hence
′
| S |=| S || P |2
′ ′
Since P is non-singular | P |6= 0. Hence | S |6= 0 (| S |= 0) if, and only if,
| S |6= 0 (| S |= 0).
This proposition has shown that the non-singularity (or yhe singularity) of
the matrices representing a symmetric bilinear form is indeed a property of the
symmetric bilinear form. This property calls for a definition a classification of
the symmetric bilinear forms and the orthogonal spaces.
Definition 5. If the matrix representative of a symmetric bilinear form is
non-singular, then the symmetric bilinear form and the orthogonal space are
called non-degenerate. Otherwise, they are called degenerate.
The final step is to reconcile this definition of non-degeneracy, based on the
matrix representatives of the form, with that already given in the main notes;
i.e. it is required to show that two definitions are equivalent.
Proposition 6. The definition of non-degeneracy given here and that given
in the main notes are equivalent.
Proof. According with the definition in the main notes, a symmetric bilinear
form is degenerate if there is a non-zero w ∈ X such that wσ is the zero mapping;
this is to say
wσ (y) ≡ (w, y) = 0 ∀ y ∈ X.
This by proposition 2,

(w, y) = (y, w) = Y T SW = 0 ∀ Y ∈ Rn ,

where S is the matrix representative of the symmetric bilinear form with respect
to a basis B, and W , and Y are the coordinate vectors of w, and y with respect
to the basis B. Therefore
SW = 0
with W 6= 0, and S is singular. Conversely, if S is singular, there exists a W 6= 0
such that
SW = 0;
hence
Y T SW = 0 ∀ Y ∈ Rn .
If w is the vector whose coordinate vector with respect to the basis B is W ,
then
(y, w) = Y T SW = 0 ∀ y ∈ X,
and w ∈ kerσ. Hence, the space is degenerate.

23
10.2 Orthogonal Maps and the Matrix Representation of
the Symmetric Bilinear Form
This section shows how the matrix representation of the symmetric bilinear form
interacts with orthogonal maps; in order to maintain the development within
reasonable limits, it is confined to orthogonal maps of an orthogonal map into
itself.
It is convenient to recall that t : X −→ X is an orthogonal map, of the
orthogonal space X, if

(t(a), t(b)) = (a, b) ∀ a, b ∈ X.

Now, we will define orthogonal mappings in a different way, by resorting to the

matrix representation of the symmetric bilinear form of the space X, and later
we will prove that both definitions are equivalent.
Definition 7. Let X be an orthogonal space, and let B be a basis of X. Let
S be the matrix representative of the symmetric bilinear form with respect to the
basis B; then a mapping t : X −→ X is orthogonal if its matrix representative
with respect to the basis B, denoted by T , satisfies

T T ST = S.

Immediately, the following question arises: Is it possible for a mapping to be

orthogonal with respect to a basis B, but not to be orthogonal to another basis
′
B ?. The next proposition solves the problem.
Proposition 8. If t is an orthogonal mapping with respect to the basis B
of the orthogonal space X, then t is orthogonal with rspect to any other basis
′
B of X.
′
Proof. Let P the transition matrix from B to B ; then the matrix represen-
′ ′
tative of t, with respect to the basis B , will be denoted by T , and it is given
by ′
T = (P T )−1 T (P T ).
Then consider
′ ′ ′
(T )T S (T ) = (P T T P −1 )(P SP T )(P T )−1 T (P T )
= P T T (P −1 P )S[P T (P T )−1 ]T (P T )
′
= P T T ST P T = P (T T ST )P T = P SP T = S .

The final proposition shows that, as expected, both definitions of an orthogonal

map coincide.
Proposition 9. The definitions of an orthogonal map given here and in tha
main notes are equivalent.
Proof. Assume that t : X −→ X is an orthogonal map according with the
definition given in the main notes. Then for all a, c ∈ X

(t(a), t(c)) = (a, c).

24
Let S be the matrix representative of the symmetric bilinear form, with respect
to a basis B. Let A and C be the coordinate vectors of a, and c with respect to
B, and T the matrix representative of t : X −→ X with respect to B. Then

(t(a), t(c)) = (T A)T S(T C) = AT T T ST C,

and
(a, c) = AT SC.
Hence, the definition of an orthogonal map leads to

AT T T ST C = AT SC ∀ A, C ∈ Rn ,

and finally
T T ST = S,
and thus t is orthogonal according to the definition given in these notes. The
proof for the other direction can be worked out following the same lines.

References
[1] Kaplansky, I., (1969), Linear Algebra and Geometry a Second Course., 2nd
ed., New York: Chelsea Publishing Company.

[2] Porteus, I. R., (1981), Topological Geometry, 2nd ed., Cambridge: Cam-
bridge University Press.

25