I.

12 Vector Spaces, Affine Spaces and Algebras 117

Affine Spaces

The abstract concept of a vector space, which plays such a fundamental role in
current mathematics, and, in particular, in modern analysis, developed from the
intuitive ‘vector calculus’ of directed arrows in our familiar three dimensional uni-
verse. A geometrical interpretation of vector space concepts is still very useful,
even in abstract situations, as we have already seen in the identification of C with
the coordinate plane. In such an interpretation we often consider vectors to be
‘points’, and certain sets of vectors to be ‘lines’ and ‘planes’, etc. To give these
concepts a solid foundation and to avoid confusion, we provide a short introduction
to affine spaces. This will allow us to use, without further comment, the language
which is most convenient for the given situation.

Let V be a K-vector space and E a nonempty set whose elements we call

points. Then E is called an affine space over V if there is a function

V ×E →E , (v, P ) → P + v

with following properties:

(AS1 ) P + 0 = P , P ∈ E.
(AS2 ) P + (v + w) = (P + v) + w, P ∈ E, v, w ∈ V .
(AS3 ) For each P, Q ∈ E there is a unique v ∈ V such that Q = P + v.
−−→
The unique vector v provided by (AS3 ) is denoted P Q. It satisfies
−−→
Q = P + PQ .
−−→
From (AS1 ) we have P P = 0, and from (AS2 ) it follows that the function
−−→
E×E →V , (P, Q) → P Q

satisfies the equation

−−→ −− → −→
P Q + QR = P R , P, Q, R ∈ E .
−−→
Since P P = 0, this implies, in particular, that
−−→ −−→
P Q = −QP , P, Q ∈ E .

Moreover, by (AS3 ), for each P ∈ E and v ∈ V there is a unique Q ∈ E such that

−−→
P Q = v, namely Q := P + v. Hence V is also called direction space of the affine
space E.
118 I Foundations

12.6 Remarks (a) The axioms (AS1 ) and (AS2 ) say that the additive group (V, +)
acts (from the right) on the set E (see Exercise 7.6). From Axiom (AS3 ) it follows
that this action has only one orbit, that is, the group acts transitively on E.
(b) For each v ∈ V ,
τv : E → E , P → P + v
is the translation of E by the vector v. It follows from (AS1 ) and (AS2 ) that the
set of translations is a subgroup of the permutation group of E.

Let E be an affine space over V . Choose a fixed point O of E, the origin.

−−→
Then the function V → E, v → O + v is bijective with inverse E → V , P → OP .
−−→
The vector OP is called the position vector of P (with respect to O).
If {b1 , . . . , bm } is a basis of V , there is
a unique m-tuple (x1 , . . . , xm ) ∈ K m such
that
m
−−→ 
OP = xj bj . ¿

j=1
¿

In this situation the numbers x1 , . . . , xm 

are called the (affine) coordinates of the ¾
point P with respect to the (affine) coordi-

¾
nate system (O; b1 , . . . , bm ). The bijective
function 

½ ½
m
E→K , P → (x1 , . . . , xm ) , (12.3)

which takes each point P ∈ E to its coordinates, is called the coordinate function
of E with respect to (O; b1 , . . . , bm ).
The dimension of an affine space is, by definition, the dimension of its direc-
tion space. A zero dimensional space contains only one point, a one dimensional
space is an (affine) line, and a two dimensional affine space is an (affine) plane.
An affine subspace of E is a set of the form P + W = { P + w ; w ∈ W } where
P ∈ E and W is a subspace of the direction space V .

12.7 Example Any K-vector space V can be considered to be an affine space

over itself. The operation of (V, +) on V is simply addition in V . In this case,
−→ = w − v for v, w ∈ V . (Here v is interpreted as a point, and the w on the left
vw
and right of the equal sign are interpreted as a point and a vector respectively!)
We choose, of course, the zero vector to be the origin.
If dim(V ) = m ∈ N× and {b1 , . . . , bm } is a basis of V , then we can identify V ,
using the coordinate function (12.3), with the standard space K m . Via (12.3),
the basis (b1 , . . . , bm ) is mapped to the standard basis e1 , . . . , em of K m . The
operations in K m lead then (at least in the case m = 2 and K = R) to the familiar
I.12 Vector Spaces, Affine Spaces and Algebras 119

‘vector calculus’ in which, for example, vector addition can be done using the
‘parallelogram rule’:

Ê
Ê
·Ê

·


½ ¼

Ê
½

¾


In the geometrical viewpoint, a vector is an arrow with head and tail at some
points, P and Q, say, of E. That is, an arrow is an ordered pair (P, Q) of points.
−−→ −−−→
Two such arrows, (P, Q) and (P
, Q
), are equal if P Q = P
Q
, that is, if there
is some v ∈ V such that P = Q + v and P
= Q
+ v, or, more geometrically, if
(P
, Q
) can be obtained from (P, Q) by some translation.

Convention Unless otherwise stated, we consider any K-vector space V to be

an affine space over itself with the zero vector as origin. Moreover we consider
K to be a vector space over itself whenever appropriate.

Because of this convention, the elements of a vector space can be called both
‘vectors’ or ‘points’ as appropriate, and the geometrical concepts ‘line’, ‘plane’
and ‘affine subspace’ make sense in any vector space.

Affine Functions

Let V and W be vector spaces over K. A function α : V → W is called affine if

there is a linear function A : V → W such that

α(v1 ) − α(v2 ) = A(v1 − v2 ) , v1 , v2 ∈ V . (12.4)

When such an A exists, it is uniquely determined by α. Indeed, setting v1 := v

and v2 := 0 we get A(v) = α(v) − α(0) for all v ∈ V . Conversely, α is uniquely
determined by A ∈ Hom(V, W ) once α(v0 ) is known for some v0 ∈ V . Indeed, for
v1 := v and v2 := v0 , it follows from (12.4) that

α(v) = α(v0 ) + A(v − v0 ) = α(v0 ) − Av0 + Av , v∈V . (12.5)

Therefore we have proved the following proposition.

120 I Foundations

12.8 Proposition Let V and W be vector spaces over K. Then α : V → W is

affine if and only if it has the form
α(v) = w + Av , v∈V , (12.6)
with w ∈ W and A ∈ Hom(V, W ). Moreover, A is uniquely determined by α, and
α is uniquely determined by A and α(0).

The interpretation of vector spaces as affine spaces makes it possible to give

geometric meaning to certain abstract objects. For the moment, this is not much
more than a language that we have transferred from our intuitions about the
three dimensional universe. In later chapters, the geometric viewpoint will become
increasingly important, even for infinite dimensional vector spaces, since it suggests
useful interpretations and possible methods of proof. Infinite dimensional vector
spaces will frequently occur in the form of function spaces, that is, as subspaces
of K X . A deep study of these spaces, indispensable for a thorough understanding
of analysis, is not within the scope of this book. This is the goal of ‘higher’ analysis,
in particular, of functional analysis.
The interpretation of finite dimensional vector spaces as affine spaces has
also an extremely important computational aspect. The introduction of coordinate
systems leads to concrete descriptions of geometric objects in terms of equations
and inequalities for the coordinates. A coordinate system is determined by the
choice of an origin and basis, and it is essential to make these choices so that the
calculations are as simple as possible. The right choice of the coordinate system
can be decisive for a successful solution of a given problem.

Polynomial Interpolation

To illustrate the above ideas, we show how interpolation questions for polynomials can
be solved easily using a clever choice of basis in Km [X]. The polynomial interpolation
problem is the following:
Given m ∈ N× , distinct x0 , . . . , xm in K and a function f : {x0 , . . . , xm } → K, find
a polynomial p ∈ Km [X] such that
p(xj ) = f (xj ) , 0≤j≤m. (12.7)

The following proposition shows that this problem has a unique solution.

12.9 Proposition There is a unique solution p := pm [f ; x0 , . . . , xm ] ∈ Km [X] of the poly-

nomial interpolation problem.

Proof The Lagrange polynomials j [x0 , . . . , xm ] ∈ Km [X] are defined by


m
X − xk
j [x0 , . . . , xm ] := , 0≤j≤m.
xj − xk
k=0
k=j
IV.1 Differentiability 303

1.3 Corollary A function f : X → E is differentiable at a if and only if it is

approximately linear at a. In this case, the approximating affine function g is
unique and given by

g: K→E , x → f (a) + f
(a)(x − a) .

Proof ‘= ⇒’ This follows directly from Theorem 1.1.

‘⇐=’ Let g : K → E be an affine function which approximates f at a. By
Proposition I.12.8, there are unique elements b, m ∈ E such that g(x) = b + mx
for all x ∈ K. Since g(a) = f (a), we have, in fact, g(x) = f (a) + m(x − a) for all
x ∈ K. The claim then follows from Theorem 1.1.

1.4 Remarks (a) Suppose that the function f : X → E is differentiable at a.

As above, define g(x) := f (a) + f
(a)(x − a) for all x ∈ K. Then the graph of g
is an affine
 line
 through a, f (a) which approximates the graph of  f near the
point a, f (a) . This line is called the tangent line to f at a, f (a) . In the case
K = R, this definition agrees with our intuitions from elementary geometry.
Ê
 






Ê¿


 

à  Ê


The expression
f (y) − f (a)
, y=a,
y−a
is called a difference quotient of f . The graph of the affine function
f (y) − f (a)
h(x) := f (a) + (x − a) , x∈K,
y−a
   
is called the secant line through a, f (a) and y, f (y) . Inthe case K = %R = E,
the differentiability of f at a means that, as y → a, the slope f (y) − f (a) (y − a)


of the secant line through
 a, f (a) and y, f (y) converges to the slope f (a) of
the tangent line at a, f (a) .
(b) Let X = J ⊆ R be an interval and E = R3 . Suppose that f (t) gives the position
of a point in space at time t ∈ J. Then |f (t) − f (t0 )|/|t − t0 | is the absolute value of
the ‘average speed’ between times t0 and t, and f˙(t0 ) represents the instantaneous
velocity of the point at the time t0 .