Anda di halaman 1dari 22

Affine Spaces

A X Athens
March 2006

Contents
1 Introduction 1

2 Definition and Examples 3

3 Chasles’ Identity 6

4 Affine Combinations 7

5 Affine Subspaces 9

6 Affine independence 14

7 Affine functions 17
7.1 Affine groups . . . . . . . . . . . . . . . . . . . . . . . . . 20

A Appendix 21

1 Introduction
Conventionally, the vector spaces R2 and R3 are taken as models of 2-
space (the plane) and 3-space. More generally, an n-space is represented
as an n-dimensional vector space. While this is essential for doing math-
ematics, it imposes certain choices which are certainly not intrinsic to
the physical understanding of 2-space or 3-space, namely a choice of ori-
gin, axes and scale. As we now see, even the more abstract notion of an
n-dimensional vector space has its shortcomings.

1
Figure 1: General change of coordinates.

To explore this further, consider the situation shown in Figure 1 which


depicts a representation of the plane in terms of coordinates (x, y) ∈ R2
with origin x = 0, y = 0. We can regard this coordinate system as a
frame {0, e1 , e2 } consisting of a choice of origin 0 and an orthogonal
choice of axes e1 = 10 for the x-direction and e2 = 01 for the y-


direction. Here lies the first point of confusion because there is no explicit
information about the physical location of the origin or the physical
directions and scale represented by the basis vectors. However, this lack
of specific physical meaning is unavoidable for an abstract framework.
The frame {0, e1 , e2 } is thus devoid of physical meaning in itself, but
can be taken as a fundamental frame in terms of which other frames can
be defined. A second frame is thus a triple {a, b1 , b2 } where a = a1 e1 +
a2 e2 is a point in the fundamental frame and the linearly independent
vectors b1 and b2 give the directions and scales of the u-axis and v-axis
respectively in terms of the fundamental frame. A given point P can now
be described by its x and y coordinates or by its u and v coordinates.
To obtain the coordinate transform, we express the fundamental basis
{e1 , e2 } in terms of the new basis {b1 , b2 }. This defines a 2 × 2 matrix

2
C such that

e1 = c1,1 b1 + c2,1 b2 and e2 = c1,2 b1 + c2,2 b2 . (1)

The relationship between our two coordinate systems is determined by

xe1 + ye2 = ub1 + vb2 + a.

Using (1), this implies

u = c1,1 (x − a1 ) + c1,2 (y − a2 )
v = c2,1 (x − a1 ) + c2,2 (y − a2 )

which can also be written as uv = C xy − Ca. Thus, the (u, v) coor-


 

dinates are obtained form the (x, y) coordinates by a linear transform


followed by a translation. The illustrates the other main weakness of
the conventional approach. A change of origin is a simple and useful
device in physical reasoning, but is awkward to deal with in the vector
space framework as a non-linear operation. For this reason, discussions
are often restricted to changes of frame with a common origin, which is
the same as a change of basis.

2 Definition and Examples


An affine space is an algebraic structure consisting of a set P of points,
a vector space V, and a mapping P + V → P satisfying the following
conditions.

(a) P + 0 = P for all P ∈ P


(b) (P + u) + v = P + (u+v) for all P ∈ P, u, v ∈ V
(c) Given P, Q ∈ P, there exists a unique v ∈ V such that P + v = Q.

The geometric interpretation of an affine space is shown in Figure 2.


The associativity axiom (b) allows the expressions such as P + v + w
to be interpreted without ambiguity. For P ∈ P and v ∈ V, we put
P − v , P + (−v). It is easy to verify the identities

(P + v) − w = P + (v − w) = P − (w−v),
(P − v) + w = P + (w − v) = P − (v − w),

3
Figure 2: Geometric interpretation of an affine space.

and
(P − v) − w = P − (v + w)
so that expressions like P − v + w can be interpreted unambiguously.
However, it should be noted that the addition or subtraction of two
points and the subtraction of a point from a vector have not been defined.
Axiom (c) can be restated as the existence of a unique mapping

P × P → V, (P, Q) 7→ P Q

such that
P + P Q = Q for all P, Q ∈ P.
Given that P, Q ∈ P and v ∈ V, the identities

(P + v)Q = P Q − v and P (Q + v) = P Q + v

are readily obtained.


It is easy to deduce that the mapping P 7→ P + v0 is bijective for any
fixed v0 ∈ V and that v 7→ P0 + v is bijective for any fixed P0 ∈ P. We
define the dimension of an affine space A to be that of the underlying
vector space V so that dim A = dim V.
Our preferred notation for a general affine space is to use capital letters
for points and lower case letters for vectors. However, when V = Rn , we
adopt the common practice of writing vectors in bold type.

Example 1 Given an n-dimensional vector space V, we can take P =


V and define addition P + V → P to be the same as addition in V.

4
This construction is known as the canonical affine structure on V.
Henceforth, whenever we refer to a vector space as an affine space, the
canonical affine structure will be implied.

Example 2 A special case of Example 1 which will provide us with


concrete examples occurs when V = Rn . In this setting, the distinction
between points and vectors is sometimes made by writing them as row
vectors and column vectors respectively so that addition is defined by
 
v1
(a1 , . . . , an ) +  ...  = (a1 + v1 , . . . , an + vn ).
 

vn

We take the view that this is not strictly necessary because we are merely
interpreting vectors in a new way. In other words, an element of Rn can
stand for a vector or a point, depending on context. It is nevertheless
convenient for presenting examples and performing computations.

Example 3 A less trivial example is provided by the line

L = {(x, y) ∈ R2 : x + y = 1}

which does not pass through the origin and is therefore not a linear
subspace of R2 . An affine space is obtained by defining addition L+R →
L as
(x, 1 − x) + v = (x + v, 1 − x − v).
We shall see later that L forms a 1-dimensional affine subspace of R2 .

Example 4 A similar example is the plane

H = {(x, y, z) ∈ R3 : x + y + z = 1}.

We obtain an affine space by defining addition H + R2 → H as


 
u
(x, y, 1 − x − y) + = (x + u, y + v, 1 − x − u − y − v).
v

As we shall see later, H forms a 2-dimensioanl affine subspace of R3 .

5
Figure 3: Chasles’ identity and the parallelogram law.

Example 5 An interesting example of an affine space is the paraboloid

P = {(x, y, x2 + y 2 ) : x, y ∈ R}.

under the operation P + R → P defined by


 
2 2 u
(x, y, x + y ) + = (x + u, y + v, (x + u)2 + (y + v)2 ).
v

In contrast to Examples 3 and 4, we shall soon see that P is not an affine


subspace of R3 .

3 Chasles’ Identity
If P, Q and R are points in an affine space then

R = Q + QR = P + P Q + QR

from which Chasles’ identity follows, namely

P Q + QR = P R.

In particular, with P = R we obtain P Q + QP = 0 so that

QP = −P Q. (2)

6
Given four points P, Q, R and S, applying Chasles’ identity yields

P Q + QR = P R = P S + SR

from which the parallelogram law follows,

P S = QR ⇔ SR = P Q.

These results are illustrated in Figure 3.

4 Affine Combinations
A linear combination of vectors is one of the fundamental concepts of
linear algebra. The analogous concept in an affine space is that of an
affine combination of points.

Lemma 6 Consider an affine space {P, V}. Let λ1 , . . . , λN be scalars


and P1 , . . . , PN be points in P.

(a) If λ1 + · · · + λN = 1 then
N
X N
X
A+ λn APn = B + λn BPn
n=1 n=1

for any two points A, B ∈ P.


(b) If λ1 + · · · + λN = 0 then
N
X N
X
λn APn = λn BPn
n=1 n=1

for any two points A, B ∈ P.

Thus, if λ1 + · · · + λN = 1 then
N
X N
X
P = λn Pn , A + λn APn
n=1 n=1

is independent of the choice of A. The point P is called an affine


combination or barycentric combination of the points P1 , . . . , PN .

7
The scalars λ1P, . . . , λN , whose sum must equal unity, are called weights.
N
We say P = n=1 λn Pn is the barycentre of the weighted points
λn Pn , (Pn , λn ). By construction, this barycentre is the unique point
X such that
XN
AX = λn APn for all A ∈ P.
n=1

Putting A = X shows that the barycentre of λ1 P1 , . . . , λn Pn is the


unique point X such that
N
X
λn XPn = 0.
n=1

For technical purposes, it is worth noting that an affine combination can


always be reduced to an expression involving only non-zero scalars. In
other words, given that λ1 + · · · + λn = 1 and P1 , . . . , Pn ∈ P, we have
X
λ1 P1 + · · · + λn Pn = λi Pi .
λi 6=0

In particular,

1P = P = P + 0Q for all P, Q ∈ P.

Proof. To establish (a), we employ Chasles’ identity to obtain


N
X N
X
A+ λn APn = A+ λn (AB + BPn )
n=1 n=1
N
X
= A + AB + λn BPn
n=1
N
X
= B+ λn BPn .
n=1

For (b) we have


N
X N
X N
X
λn APn = λn (AB + BPn ) = λn BPn
n=1 n=1 n=1

as required.

8
Example 7 Consider the canonical affine structure on a vector space V.
If v1 , . . . , vN ∈ V and λ1 + · · · + λN = 1 then
N
X
a+ λn (vn − a) = λ1 v1 + · · · + λn vn
n=1

for all a ∈ V. This shows that affine combination are the same as linear
combinations in this setting.

5 Affine Subspaces
A linear subspace of a vector space V is defined as a non-empty subset of
V which is closed under linear combinations. The corresponding notion
in an affine space is that of an affine subspace, which is a subset that is
closed under affine combinations.
Let A = {P, V} be an affine space. Then a subset S of P is an affine
subspace of A if λ1 P1 + · · · + λN PN ∈ S whenever P1 , . . . , PN ∈ S and
λ1 + · · · + λN = 1. It follows immediately from this definition that the
empty set is a trivial affine subspace and that any intersection of affine
subspaces is an affine subspace.
By definition, an affine subspace S of A must contain λP + (1 − λ)Q
whenever λ ∈ R and P, Q ∈ S. In fact, this condition is sufficient to
ensure that S is an affine subspace.

Lemma 8 Let A = {P, V} be an affine space and S ⊂ P. If

(1 − λ)P + λQ ∈ S for all λ ∈ R, P, Q ∈ S (3)

then S is an affine subspace of A.


The condition (3) can of course be written as

P + λP Q ∈ S for all λ ∈ R, P, Q ∈ S

If P, Q ∈ S and P 6= Q then LP,Q = {P + λP Q : λ ∈ R} is the line


passing through P and Q. This result states that S is an affine subspace
if and only if LP,Q ⊂ S whenever P, Q ∈ S.

Proof. As an induction hypothesis, let n ≥ 2 and suppose that S con-


tains all affine combinations of n or fewer points in S. Let S0 , . . . , Sn ∈ S

9
and λ0 + · · · + λn = 1 with λ0 6= 1. We must show that

P = λ0 S0 + · · · + λn Sn ∈ S.

Our induction hypothesis ensures that


λ1 λn
S= S1 + · · · + Sn ∈ S
1 − λ0 1 − λ0
and
P = λ0 S0 + (1 − λ0 )S ∈ S
as required.

Example 9 Let a, b, c ∈ R with a 6= 0 or b 6= 0. Then it is easy to verify


that
L = {(x, y) ∈ R2 : ax + by = c} (4)
is an affine subspace of R2 . Two of these lines with a = 1 = b are shown
in Figure 4. One has c = 0 and the other has c = 1. Since these two
lines are parallel, they have no point of intersection. This shows why it
is convenient to admit the empty set as an affine subspace. Otherwise
an intersection of affine subspaces would not automatically be an affine
subspace. This example also highlights a difference between linear and
affine subspaces. The line (4) is a linear subspace of R2 if and only if
c = 0. One of the lines shown in Figure 4 is a linear subspace of R2 but
the other is not.

Example 10 Let U be a linear subspace of a vector space V. Given


v0 ∈ V, it is easy to verify that U + v0 is an affine subspace of V.
Examples 3 and 4 are both special cases of this.

Theorem 11 Let A = {P, V} be an affine space. If S ⊂ P then the set


hSi of all affine combinations

λ1 S1 + · · · + λN SN with S1 , . . . , SN ∈ S (5)

is an affine subspace of A.
By definition of an affine subspace, it is clear that hSi is the smallest
one containing S, i.e. the intersection of all affine subspaces T of A such
that S ⊂ T . We call hSi the affine subspace generated by S. We also

10
Figure 4: Affine subspaces of R2 .

denote the affine subspace generated by a finite set of points P1 , . . . , Pn


as hP1 , . . . , Pn i.

Proof. It is easy to verify that any affine combination of points of the


form (5) is of the same form.

Lemma 12 Let A = {P, V} be an affine space.

(a) If S is an affine subspace of A and A ∈ S then

VA , {AP : P ∈ S} = {P Q : P, Q ∈ S} (6)

is a linear subspace of V and S = A + VA = S + VA .


(b) If U is a linear subspace of V and A ∈ P then A + U is an affine
subspace of P.

Note that if S ⊂ P and U ⊂ V then

S + U , {P + u : P ∈ S and u ∈ U}.

11
If S is an affine subspace of A then (a) shows that {S, VA } is an affine
space in itself with respect to the operation S + VA → S induced by
P + V → P. We call the vector space VA the direction of the affine
subspace S and take dim VA to be the dimension of S. It follows from (a)
that the affine subspace generated by a finite set of points P1 , . . . , Pn
must have dimension ≤ n − 1.

Proof. Suppose S is an affine subspace of A and A ∈ S. If P, Q ∈ S


and t ∈ R then

R = A + AP + tAQ = A − tAA + AP + tAQ

is an affine combination of the points A, P and Q so that R ∈ S.


Consequently
AR = AP + tAQ ∈ VA
proving that VA is a linear subspace of V. Furthermore, if P, Q ∈ S then

P Q = P A + AQ = AQ − AP ∈ VA

which gives the equality in (6). By definition of VA , we have

A + VA ⊂ S ⊂ S + VA .

Moreover, if P, Q ∈ S then

P + AQ = A + AP + AQ ∈ A + VA .

This completes the proof of (a). The proof of (b) is left as an easy
exercise for the reader.
Statement (b) of Lemma 12 shows that the line and plane of Exam-
ples 3 and 4 are affine subspaces of R2 and R3 respectively.

Example 13 In Example 5, we saw that the paraboloid

P = {(x, y, x2 + y 2 ) : x, y ∈ R}

forms a 2-dimensional affine space under the operation


 
2 2 u
(x, y, x + y ) + = (x + u, y + v, (x + u)2 + (y + v)2 ).
v

12
Since P contains the point (0, 0, 0) and is not a linear subspace of R3 , it
cannot be an affine subspace of R3 by part (a) of Lemma 12. Fixing y,
we obtain the one-dimensional affine subspace
 
2 2 2 1
S = {(x, y0 , x + y0 ) : x ∈ R} = (0, y0 , y0 ) + R
0
which is a parabola.
An affine subspace of dimension one is called a line and an affine
subspace of dimension two is called a plane. Given an affine space
A = {P, V} and a subset S of P, the points contained in S are said to
be
(a) collinear if S is contained in a 1-dimensional affine subspace of A
(b) coplanar if S is contained in a 2-dimensional affine subspace of A.

Theorem 14 Let {P, V} be an affine space and A, B, C, D ∈ P.


(a) The points A, B and C are collinear if and only if the vectors AB
and AC are linearly dependent.
(b) The points A, B, C and D are coplanar if and only if the vectors
AB, AC and AD are linearly dependent.
Proof. Suppose A, B and C are collinear. If A = B then AB = 0. If
AB 6= 0 then
hA, Bi = {A + tAB : t ∈ R}
is the line passing through A and B and so we must have AC = tAB
for some t ∈ R which shows that the vectors AB and AC are linearly
dependent.
Assume now that AB and AC are linearly dependent. If AB = 0 then
A = B whence A, B, C ∈ hA, Ci. If AB 6= 0 then AC = tAB for some
t ∈ R so that C ∈ hA, Bi. This establishes (a) and the proof of (b) is
similar.
Let A = {P, V} be an affine space. If V = U ⊕ Rv for some v ∈ V
then any affine subspace of the form A + U called a hyperplane. We
say that two affine subspaces are parallel if their directions are equal.
Thus, if S and T are parallel affine subspaces of A then there exist
points A, B ∈ P and a linear subspace U of V such that S = A + U and
T = B + U. Parallel affine subspaces are obtained by translation in the
sense that T = S + AB.

13
6 Affine independence
Lemma 15 Let (P, V) be an affine space and Pj ∈ P for each j in
an index set J. If the family of vectors (Pi Pj : i 6= j ∈ J) is linearly
independent for some i ∈ J then the same is true for all i ∈ J.

The family of points (Pj : j ∈ J) is said to be affine independent if the


family of vectors (Pi Pj : i 6= j ∈ J) is linearly independent for some i ∈ J
and affine dependent otherwise. We observe that if (Pi Pj : i 6= j ∈ J)
is linearly dependent for some i ∈ J then the same is true for all i ∈ J.

Proof. Suppose i ∈ J and (Pi Pj : i 6= j ∈ J) is linearly independent.


Let i 6= k ∈ J and X
λj Pk Pj = 0.
k6=j∈J

Since Pk Pj = Pi Pj − Pi Pk , we have
X X X
λ j Pk Pj = λj Pi Pj − λ j Pi Pk
j6=k j6=k j6=k
 
X X 
= λ j Pi Pj − λj Pi Pk
 
j6=i,k j6=k

P
so that λj = 0 for i, k 6= j ∈ J and j6=k λj = λi = 0 as required.

Theorem 16 Let (P, V) be an affine space and P0 , . . . , Pn ∈ P. Then


each P ∈ hP0 , . . . , Pn i has a representation

P = λ0 P0 + · · · + λn Pn with λ0 + · · · + λn = 1. (7)

This representation is unique if the points P0 , . . . , Pn are affine indepen-


dent and non-unique otherwise.

Proof. Let λ1 , . . . , λn ∈ R and put λ0 = 1 − λ1 − · · · − λn . Since

λ0 P0 + · · · + λn Pn = P0 + λ1 P0 P1 + · · · + λn P0 Pn

the representation (7) is unique if and only if the vectors P0 P1 , . . . , P0 Pn


are linearly independent. By definition, the latter condition is equivalent
to the affine independence of the points P0 , . . . , Pn .

14
Let A = (P, V) be an affine space and J be an index set with 0 ∈ J.
A family of points F = (Aj : j ∈ J) is called an affine frame for A if
the family of vectors (A0 Aj : 0 6= j ∈ J) forms a basis of V. If F is an
affine frame for A then the family (A0 , A0 Aj : 0 6= j ∈ J) is called an
affine frame with origin A0 . Thus, if dim V = n then an affine frame
of A must consist of n + 1 points A0 , . . . , An and any point P ∈ P can
be written uniquely in the form

P = A0 + x1 b1 + · · · + xn bn
= x 0 A0 + x 1 A1 + · · · + x n An

where bj = A0 Aj and x0 = 1−x1 −· · ·−xn . The n scalars x1 , . . . , xn are


called the coordinates of P with respect to the frame (A0 , b1 , . . . , bn ).
The n+1 scalars x0 , . . . , xn are known as the barycentric coordinates
of P with respect to the frame (A0 , . . . , An ).

Lemma 17 Let (P, V) be an affine space.

(a) A finite sequence of points P0 , . . . , Pn ∈ P is affine dependent if


and only if there exist scalars λ0 , . . . , λn , not all zero, such that
λ0 + · · · + λn = 0 and

λ0 P P0 + · · · + λn P Pn = 0 for all P ∈ P.

(b) A family F of points in P is affine dependent if and only some finite


subfamily of F is affine dependent.

Proof. It follows from Lemma 15 and the subsequent definition that


the points P0 , . . . , Pn are affine dependent if and only if the vectors
P0 P1 , . . . , P0 Pn are linearly dependent. The latter condition is equivalent
to the existence of scalars λ1 , . . . , λn , not all zero, such that

λ1 P0 P1 + · · · + λn P0 Pn = 0. (8)

Given P ∈ P and λ1 , . . . , λn , we have


n
X n
X n
X
λj P0 Pj = λj (P Pj − P P0 ) = λj P Pj
j=1 j=1 j=0

where λ0 = −λ1 · · · − λn . Comparing the last expression with (8) es-


tablishes (a). For the proof of (b), we recall that an infinite family G

15
of vectors is defined to be linearly independent if every finite subfamily
of G is linearly independent and is therefore linearly dependent if some
finite subfamily is linearly dependent.
Two points A, B ∈ P are affine independent if and only if AB 6= 0,
i.e. A 6= B. If A 6= B then

hA, Bi = {(1 − λ)A + λB : λ ∈ R} = {A + λAB : λ ∈ R}

is the line passing through A and B.


Three points A, B, C ∈ P are affine independent if and only if the
vectors AB and AC are linearly independent. If A, B and C are affine
independent then

hA, B, Ci = {(1 − λ − µ)A + λB + µC : λ, µ ∈ R}


= {A + λAB + µAC}

is the plane containing A, B and C. By (a) of Theorem 14, three points


are affine dependent if and only if then are collinear.
Four points A, B, C, D ∈ P are affine independent if and only if the
vectors AB, AC and AD are linearly independent. Part (b) of Theo-
rem 14 shows that four points are affine dependent if and only if they
are coplanar.
Let (P, V) be an affine space. If P0 , . . . , Pn ∈ P then any point P ∈ P
given by
P = λ0 P0 + · · · + λn Pn
for some λ0 , . . . , λn ∈ R with

λ0 + · · · + λn = 1 and λ0 , . . . , λn ≥ 0

is called a convex combination of P0 , . . . , Pn . A set C ⊂ P is called


a convex set if C contains all convex combinations of points in C. If
S ⊂ P then the set CH(S) of all convex combinations of points in S is
known as the convex hull of S.

Theorem 18 Let (P, V) be an affine space.

(a) A subset C of P is convex if and only if

(1 − λ)A + λB = A + λAB ∈ C (9)

whenever A, B ∈ C and 0 < λ < 1.

16
(b) Any intersection of convex sets is convex.

(c) If S ⊂ P then CH(S) is convex.

Statements (b) and (c) show that CH(S) is the smallest convex set
containing S.

Proof. To establish the non-trivial part of (a), assume as an induction


hypothesis that n ≥ 2 and C contains all convex combinations of k points
in C when 2 ≤ k ≤ n. Suppose

P = λ0 C0 + · · · + λn Cn

with C0 , . . . , Cn ∈ C,

λ0 + · · · + λn = 1 and λ0 , . . . , λn ≥ 0.

It may be assumed that 0 < λ0 < 1 without loss of generality. Then it


follows from the induction hypothesis that

λ1 λn
C= C1 + · · · + Cn ∈ C
1 − λ0 1 − λ0

and
P = λ0 C0 + (1 − λ0 )C ∈ C
as required. The proof of (b) and (c) is left as an easy exercise for the
reader.

7 Affine functions
Let (P, V) and (Q, W) be affine spaces. A map f : P → Q is called an
affine function if

f ((1 − λ)P + λQ) = (1 − λ)f (P ) + λf (Q)

for all λ ∈ R and for all P, Q ∈ P.

Theorem 19 Suppose (P, V) and (Q, W) are affine spaces and f : P →


Q is an affine function.

17
(a) If P1 , . . . , Pn ∈ P and λ1 + · · · + λn = 1 then

f (λ1 P1 + · · · + λn Pn ) = λ1 f (P1 ) + · · · + λn f (Pn ). (10)

(b) If S is an affine subspace of (P, V) then f (S) is an affine subspace


of (Q, W).

(c) If T is an affine subspace of (Q, W) then f −1 (T ) is an affine sub-


space of (P, V).

(d) If f is injective then f −1 is an affine mapping of f (P) onto P.

We refer to a mapping f : P → Q which is both affine and bijective


as an affine bijection. We also use the term affine transform when
P = Q and V = W. If f is an affine bijection then f −1 : P → Q is affine
by (d). An affine bijection is thus an isomorphism for affine spaces.

Proof. The definition of an affine function ensures that (10) holds for
n = 2. Assume now that the relation holds for some positive integer
n ≥ 2. Let P0 , . . . , Pn ∈ P and λ0 + · · · + λn = 1 with λ0 6= 1. Putting

λ1 λn
P = P1 + · · · + Pn
1 − λ0 1 − λ0
we have
λ0 P0 + · · · + λn Pn = λ0 P0 + (1 − λ0 )P
so that

f (λ0 P0 + · · · + λn Pn ) = λ0 f (P0 ) + (1 − λ0 )f (P )
= λ0 f (P0 ) + λ1 f (P1 ) + · · · + λn f (Pn )

thus establishing (a). The proof of (b) and (c) is left as an easy exercise
for the reader.

Theorem 20 Let (P, V) and (Q, W) be affine spaces.

(a) If L : V → W is a linear map, A ∈ P and B ∈ Q then an affine


map f : P → Q is given by

f (P ) = B + L(AP ).

18
(b) If f : P → Q is an affine map and A ∈ P then a linear map
Lf : V → W is given by

Lf (v) = f (A)f (A + v). (11)

This linear map is independent of the choice of A.

The map Lf in (b) is referred to as the linear map associated with f .

Proof. The proof of (a) is left as an easy exercise for the reader. To
establish (b), fix A ∈ P. If λ ∈ R and v ∈ V then

A + λv = (1 − λ)A + λ(A + v)

so that

Lf (λv) = f (A){(1 − λ)f (A) + λf (A + v)}


= λf (A)f (A + v) = λLf (v).

If u, v ∈ V then
1 1
A+u+v = (A + 2u) + (A + 2v)
2 2
whence
 
1 1
Lf (u + v) = f (A) f (A + 2u) + f (A + 2v)
2 2
1 1
= f (A)f (A + 2u) + f (A)f (A + 2v)
2 2
1 1
= Lf (2u) + Lf (2v) = Lf (u) + Lf (v).
2 2
To complete the proof of (b), we observe that if u, v ∈ V then

f (A + u)f (A + v) = f (A)f (A + v) − f (A)f (A + u)


= Lf (v) − Lf (u).

Hence, if B = A + b for some b ∈ V and v ∈ V then

f (B)f (B + v) = f (A + b)f (A + b + v)
= Lf (b + v) − Lf (b) = Lf (v)

19
as required.
An affine function f : P → P is called a translation if Lf is the
identity map. It follows from (b) of Theorem 20 that f is a translation
if and only if, for any fixed point A ∈ P, we have

f (A + v) = f (A) + v for all v ∈ V.

The easy proof of the following results is left as an exercise for the reader.

Proposition 21 Let (P, V) and (Q, W) be affine spaces and f : P → Q


be an affine function.

(a) f is injective if and only if Lf is injective.


(b) f (P) = Q if and only if Lf (V) = W.
(c) f is constant if and only if Lf = 0.

7.1 Affine groups


The following result is easy to establish.

Lemma 22 Let P, Q and R be affine spaces. If

f :P→Q and g:Q→R

are affine functions then so is g ◦ f and we have Lg◦f = Lg ◦ Lf . If f is


an affine bijection then Lf is a linear isomorphism and Lf −1 = L−1f .

Given an affine space A = (P, V), the set AG(A) of all affine transforms
of P forms a subgroup of the permutation group SP and is known as
the affine group of A. The set GL(V) of all linear automorphisms of V
forms a subgroup of SV . Lemma 22 shows that the mapping

L : AG(A) → GL(V) given by f 7→ Lf

is a homomorphism of the affine group AG(A) onto the general linear


group GL(V). By the definition following the proof of Theorem 20, the
kernel of this homomorphism

ker L = {f ∈ AG(A) : Lf = IV }

is precisely the set of all translations of A.

20
We denote the multiplicative group of non-zero real numbers by R∗ .
The linear mappings R∗ IV , called dilations, form a subgroup of GL(V)
which is isomorphic to R∗ under the homomorphism λ 7→ λIV . The
members of the affine subgroup DIL(A) , L−1 (R∗ IV ) are called affine
dilations. Given A ∈ P and λ ∈ R, the affine map

HA,λ : P → P given by HA,λ (P ) = A + λAP

is called a dilation or homothety. This dilation is said to be centred


at A and to have ratio λ. We make the following observations.

(a) If λ 6= 0 then P 7→ HA,λ (P ) is an affine transform of P.

(b) A is a fixed point of HA,λ and is the unique fixed point if λ 6= 0.


(c) If B 6= A then λ 7→ HA,λ (B) is an affine bijection of R onto the line
passing through A and B.

If V is a normed space then a metric d on P is given by

d(P, Q) = kP Qk .

It is easy to verify that the affine dilation HA,λ satisfies

HA,λ (P )HA,λ (Q) = λP Q for all P, Q ∈ P. (12)

If λ = 0 then HA,λ is just the constant A. For λ 6= 0, the geometric


interpretation of (12) is that HA,λ scales the distance between any two
points by the associated ratio λ, while leaving the direction of any affine
subspace of A unchanged. For example, if ∆ is a triangle then HA,2/3 (∆)
is a triangle two thirds the size of ∆ as shown in Figure 5.

A Appendix

21
Figure 5: Affine dilation of a triangle with fixed point A and ratio λ = 2/3.

22

Anda mungkin juga menyukai