Isometries of R

n
S. Kumaresan
Dept. of Math. & Stat.
University of Hyderabad
Hyderabad 500 046
[email protected]
September 20, 2004
Let (X, d) and (y, d) be metric spaces. A map f : X Y is said to be an isometry if
d(f(x
1
), f(x
2
)) = d(x
1
, x
2
) for all x
1
, x
2
X. Note that an isometry is always one-one but in
general not onto. For instance, consider f : [1, ) [1, ) given by f(x) = x + 1. If f and
g are isometries of X to itself, then g f is also an isometry. The set of surjective isometries
of a metric space form a group under the composition.
We consider R
n
with the Euclidean inner product x, y =

n
i=1
x
i
y
i
. We then have the
notion of norm or length of a vector x :=

x, x. It is well-known that d(x, y) := x y

denes a metric on X. Thus (R
n
, d) becomes a metric space. The aim of this article is to
give a complete description of all isometries of R
n
and look a little more geometrically into
the isometries of R
3
.
First a bit of convention: We consider R
n
as the vector space of column vectors, that is,
n1 real matrices. Given an nn matrix A, we have a linear map on R
n
given by x Ax.
In the sequel, we shall not distinguish between the matrix A and the associated linear map.
I am sure that the context will make it clear what we are referring to.
Let us rst of all look at some examples of isometries. For a xed v R
n
, consider the
translation T
v
: R
n
R
n
given by T
v
(x) := x + v. Then T
v
is an isometry of R
n
:
d(T
v
x, T
v
y) = T
v
x T
v
y = (x + v) (y + v) = x y = d(x, y).
We now describe another important class of isometries of R
n
. A linear map f : R
n
R
n
is said to be orthogonal if it preserves inner products: for every pair x, y R
n
, we have
f(x), f(y) = x, y .
The following result is well-known. For a proof, I refer the reader to my book on linear
algebra.
Theorem 1. For a linear map f : R
n
R
n
the following are equivalent:
1. f is orthogonal.
2. f(x) = x for all x R
n
.
3. f takes an orthonormal basis of R
n
to an orthonormal basis.
We now look at a very special class of orthogonal maps.
Fix a unit vector u R
n
. Let W := (Ru)

:= {x R
n
: x, u = 0}. Then W is a
vector subspace of dimension n 1. This can be seen as follows. The map f
u
: R
n
R
1
given by f
u
(x) := x, u is linear. It is nonzero, since f
u
(u) = 1. Hence the image of f
u
is a
nonzero vector subspace and hence all of R. Also, we observe that W = ker f
u
. Hence by the
rank-nullity theorem,
n = dimR
n
= dimker f
u
+ dimImf
u
.
It follows that W is an n1 dimensional vector subspace of R
n
. We thus have an orthogonal
decomposition R
n
= W Ru. We use W as a mirror and reect across it. Thus any vector in
W is mapped to itself whereas the vector u is mapped to u. Thus the reection
W
: R
n

R
n
is given by (x) = w tu, where x = w + tu, t R. This map is clearly an orthogonal
linear map. For, if {w
1
, . . . , w
n1
} is an orthonormal basis of W, then {w
1
, . . . , w
n1
, u}
is an orthonormal basis of R
n
. The linear map
W
carries this orthonormal basis to the
orthonormal basis {w
1
, . . . , w
n1
, u} and hence is orthogonal. We have another description
of this map as follows:

W
(x) := x 2 x, u u. (1)
The advantage of this expression is that it is basis-free. We also observe that the expression
remains the same if we replace u by u.
We shall look at the case when n = 2 in detail. We shall derive a matrix representation
of
W
from (1). Consider the one dimensional subspace R

cos t
sint

. Let u =

sint
cos t

. Then

W
(e
1
) =

1
0

1
0

sint
cos t

sint
cos t

1 2 sin
2
t
2 sint cos t

cos 2t
sin2t

W
(e
2
) =

0
1

0
1

sint
cos t

sint
cos t

sin2t
cos 2t

.
Thus the reection about the line is given by
W
=

cos 2t sin2t
sin2t cos 2t

.
Remark 2. Reections generate the group of orthogonal linear maps. (We shall not prove
this.) Thus, they are the building blocks of orthogonal linear maps. If we observe that
2
= 1
for any reection, then the analogy between the transpositions and reections is striking.
A surprising result is
Theorem 3. Let f : R
n
R
n
be an isometry with f(0) = 0. Then f is an orthogonal linear
map.
A proof of this can also be found in my book. However, we indicate a slightly dierent
proof found in an unpublished note by Artin.
Lemma 4. Let x, y R. Assume that x, x = x, y = y, y. Then x = y.
Proof. We compute the length square of x y:
x y, x y = x, x 2 x, y + y, y = 0,
in view of our hypothesis. Hence x = y.
2
Lemma 5. An isometry of R
n
that xes the origin preserves the inner products.
Proof. Let f : R
n
R
n
be an isometry such that f(0) = 0. We need to prove that
f(x), f(y) = x, y for all x, y R
n
. Since d(f(x), f(y)) = d(x, y), we have f(x) f(y) =
x y. Taking squares, we get
f(x) f(y), f(x) f(y) = x y, x y , for all x, y R
n
. (2)
Since f(0) = 0, setting y = 0 in (2), we get f(x), f(x) = x, x. Similarly, f(y), f(y) =
y, y. Now, if we expand both sides of (2) and cancelling equal terms such as x, x and
y, y, we get the desired result.
We now prove Theorem 3.
Proof of Theorem 3. We need only show that f is linear. Let x, y R
n
. Let z = x + y.
We shall show rst that f(z) = f(x) + f(y). In view of Lemma 4, it suces to show that
f(z), f(z) = f(z), f(x) + f(y) = f(x) + f(y), f(x) + f(y) .
Expanding the last two terms, we need to show that
f(z), f(z) = f(z), f(x) + f(z), f(y) = f(x), f(x) + 2 f(x), f(y) + f(y), f(y) . (3)
Since f preserves inner products, (3) holds same i
z, z = z, x + z, y = x, x + 2 x, y + y, y , (4)
holds. Since z = x + y, (4) is clearly true!
Similarly, if we take y = ax, a R, then we need to show that f(y) = af(x). In view of
Lemma 4, it is enough to show that
f(ax), f(ax) = f(ax), af(x) = af(x), af(x) .
Since f is inner product preserving, it suces to show that
ax, ax = a ax, x = a
2
x, x ,
which is true.
As immediate consequences of Theorem 3, we have
Corollary 6. (i) Let f : R
n
R
n
be an isometry. Then f = T
v
A where v = f(0).
(ii) Any isometry of R
n
is onto.
Proof. Consider g := T
v
f where v = f(0). Then clearly, g is an isometry of R
n
such that
g(0) = 0. Hence g is an orthogonal linear map, say, A. Hence f = T
v
A. Thus (i) is proved.
Since any translation and any orthogonal linear map are onto, so is an isometry.
We now give a complete list of all orthogonal maps of R
2
. Let A: R
2
R
2
be an
orthogonal. If we represent A with respect to the standard orthonormal basis {e
1
, e
2
} as a
matrix, then A is either of the form

cos t sint
sint cos t

or of the form

cos t sint
sint cos t

.
3
In the rst case, we say that it is a rotation by an angle t in the anticlockwise direction. In
the latter case, it is reection with respect to the line R

cos(t/2)
sin(t/2)

, as seen earlier.
Thus we have understood all the orthogonal maps of R
2
. We now wish to do the same in
R
3
.
Denition 7. Given a two dimensional vector subspace W of R
3
, we dene a rotation in the
plane W as follows. Fix an orthonormal basis {w
1
, w
2
} of R
3
. Fix a unit vector u such that
u W. We consider the linear map given by
R
W,t
(w
1
) = cos t w
1
+ sint w
2
R
w,t
(w
2
) = sint w
1
+ cos t w
2
R
w,t
(u) = u.
Clearly, R
W,t
is orthogonal. Also, with respect to the orthonormal basis {w
1
, w
2
, u}, it is
represented by

cos t sint 0
sint cos t 0
0 0 1

.
We shall refer to R
W,t
as the rotation in the plane W with Ru as the axis of rotation. Note
that det(R
W,t
) = 1.
Proposition 8. (i) Let A be a 3 3 orthogonal matrix. Assume that det A = 1. Then 1 is
an eigenvalue of A.
(ii) Let A be a 3 3 orthogonal matrix. Assume that det A = 1. Then A is a rotation.
Proof. (i). Consider the following chain of equations:
det(A I) = det(A AA
t
) = det(A(I A
t
))
= det(A) det(I A
t
) = det((I A
t
)
t
) = det(I A).
For any n n matrix B, we have det(B) = (1)
n
det B. Hence, we see that det(A I) =
det(I A) = det(A I). We conclude that det(A I) = 0. From this, (i) follows.
We now prove (ii). By (i), there exists a unit vector u R
3
such that Au = u. Let
W := (Ru)

. Then W is a two dimensional vector subspace. We claim that Aw W for any

w W. This is seen as follows:
Aw, u = Aw, Au = w, u = 0.
Thus A |
W
: W W is an orthogonal linear map. If we choose an orthonormal basis {w
1
, w
2
}
of W, then A can be represented with respect to the orthonormal basis {w
1
, w
2
, u} either as

cos t sint 0
sint cos t 0
0 0 1

or as

cos t sint 0
sint cos t 0
0 0 1

.
The second case cannot occur as the determinant is -1, contradicting our hypothesis. Hence
(ii) follows.
4
Theorem 9. Let A be any 3 3 orthogonal matrix. Then either A is a rotation or is a
rotation followed by a reection.
Proof. If det A = 1, then we know that it is a rotation. So we assume that det A = 1.
Then consider the reection matrix corresponding to the xy plane: = diag(1 1 1). Then
B = A has determinant 1 and hence is a rotation. Since A = B, the theorem follows.
Remark 10. Let A be an orthogonal 33 matrix with det A = 1. How can we decide whether
A is a pure reection or is a rotation followed by a reection? If it is a pure reection, then
its eigenvalues are +1, +1, 1. In the other case, either all eigenvalues are 1 or it has only
one real eigenvalue, namely 1.
This article is based on a lecture given in a Workshop for College Teachers, organized
by Bombay Mathematics Colloquium at Bhavans College on September 15, 2004. I thank
Professor Dhvanita Rao for the invitation and the audience for an enthusiastic response.
Reference
S. Kumaresan, Linear AlgebraA Geometric Approach, Prentice-Hall of India, 2000.
5