Geometry of Vector Spaces

Fall 2014 MATH430

In these notes we show that it is possible to do geometry in vector spaces as well,

that is similar to plane geometry. We start by giving the definition of an abstract vector
space:
Definition 1. (V, +.·) is a real vector space if for any u, v, w ∈ V and r, p ∈ R the
following hold:
• u + v = v + u,

• u + (v + w) = (u + v) + w,

• there exists a unique vector 0 ∈ V such that 0 + u = u and u + (−u) = 0,

• 1 · u = u,

• r · (u + v) = r · u + r · v,

• (r + p) · u = r · u + p · u,

• r · (p · u) = (r · p) · u.
The usual example of a vector space is the plane (R2 , +, ·) or the space (R3 , +, ·) with
vector addition and scalar multiplication. A less conventional example is the space of
continuous functions on the unit interval (C[0, 1], +, ·), where vector addition is addition
of functions and scalar multiplication is multiplying functions with real numbers:

(f + g)(x) = f (x) + g(x),

(rf )(x) = r · f (x),

where f, g ∈ C[0, 1], r ∈ R, x ∈ [0, 1]. The zero vector in this space is just the identically
zero function.
Remark 1. One should think of scalar multiplication with a positive number as dilation
(zooming in/out) and multiplication with −1 is just reflection in the zero vector 0.
It turns out that the notion of parallelism can be defined in any vector space (V, +, ·):
Definition 2. Two vectors u, v ∈ V are parallel if there exists a non-zero number r ∈ R
such that u = r · v.
Definition 3. Given vectors a, b, u, v ∈ V , the segments [ab] and [uv] are parallel if the
vector b − a is parallel to the vector v − u, as illustrated in the figure below.

1
 On the surface, vector spaces have fairly little structure. However it turns out that
this structure is more then enough to prove vector space versions of classical theorems
from the plane involving parallelism:
Theorem 1 (Vector Thales). Suppose (V, +, ·) is a vector space and s, v, t, w ∈ V are
vectors such that s is parallel with v and t is parallel with w and v is not parallel with
w. If w − v is parallel with t − s then v = a · s and w = a · t for some number a.

Proof. We have
s = bv,
and
t = cw
for some b, c ∈ R because s is parallel with v and t is parallel with w. To prove the
theorem, we need to argue that b = c. As w − v is parallel with t − s we also have

d(w − v) = (t − s).

We can replace s and t in this last identity with bv and cw to obtain

d(w − v) = cw − bv.

After organizing terms we end up with

(d − c)w = (d − b)v.

As w and v are not parallel we need to have d − c = 0 and d − b = 0 above. In particular

b = c, which is what we needed to prove.

2
Theorem 2 (Vector Pappus). Suppose (V, +, ·) is a vector space and r, s, t, u, v, w ∈ V
lie alternately on two lines through the zero vector 0 according to the figure below. Suppose
u − v is parallel to s − r and t − s is parallel to v − w. Then u − t is parallel to w − r.

Proof. We want to show that there exists a non-zero number h such that u−t = h(w−r).
Because u − v is parallel to s − r by the vector Thales theorem we have

u = as and v = ar.

As t − s and v − w are also parallel by the same reason we have

s = bw and t = bv.

Putting these identities together we obtain u = abw and t = bar. Hence u−t = ab(w−r),
proving that u − t is parallel with w − r.

Lines, midpoints and centroids

We now describe the equations of a lines in vector spaces. As we will see, the concept of
line from the plane goes through without any changes.
Given two vectors v, d ∈ V the parametric equation of a line l through the ”point” v
in the ”direction” of d is given by

l : p = v + td, t ∈ R, (1)

where p is an arbitrary vector on the line l:

Similarly, the equation of the line h through the ”points” v, w ∈ V is given by

h : p = v + t(w − v) = tw + (1 − t)v, t ∈ R. (2)

3
 If we put t = 1/2 in the equation of the line h above, then the vector p is just the
midpoint of the segment [vw]:
1
p = (v + w).
2
In general, given vectors v1 , v2 , . . . , vn ∈ V the point
1
p = (v1 + · · · + vn ).
n
is called the centroid of v1 , v2 , . . . vn .
Problem 1 (Concurrence of medians). Suppose (V, +, ·) is a vector space and v1 , v2 , v3 ∈
V . Prove using vector geometry that the medians of the triangle v1 v2 v3 intersect each
other in the centroid 31 (v1 + v2 + v3 ).

Solution. By the discussion above the midpoints of the segments [v1 v2 ], [v2 v3 ], [v3 v1 ] are
the vectors (v1 + v2 )/2, (v2 + v3 )/2, (v1 + v3 )/2 respectively.
We need to show that the vector 31 (v1 + v2 + v3 ) is on the line joining v3 and v1 +v 2
2
.
According to the equation of a line through two given points in (2), we need to find t ∈ R
such that
1 v1 + v2
(v1 + v2 + v3 ) = tv3 + (1 − t) .
3 2
Isolating the coefficients of the vectors v1 , v2 , v3 in the above identity we end up with
1 1 − t 1 1 − t 1 
− v1 + − v2 + − t v3 = 0.
3 2 3 2 3
We notice that the value t = 1/3 makes all the coefficients become zero. One can
similarly prove that 13 (v1 + v2 + v3 ) is on the line joining v2 and v1 +v2
3
and also on the
v1 +v2
line joining v3 and 2 finishing the proof.
Without a doubt the above argument is very general and ’clean’. As a consequence
of this last proof, we also obtain that the centroid v1 +v32 +v3 divides each median in
proportions 13 : 23 .

4
The inner/scalar product
In the preceding section we learned that in a vector space (V, +, ·) one can introduce
parallelism and the concept of a line can also be defined. This two notions where already
very useful. However there is no natural way to introduce angles and distances. As we
will see shortly, all of this is possible if on has an inner/scalar product in a vector space:
Definition 4. Given a vector space (V, +, ·) an inner/scalar product is a function h·, ·i :
V × V → R that satisfies the following conditions for u, v, w ∈ V and r ∈ R:
• (symmetry) hu, vi = hv, ui,

• (linearity) hu + w, vi = hu, vi + hw, vi and hr · u, vi = rhu, vi,

• (positivity) hu, ui ≥ 0 and if hu, ui = 0 then u = 0.

A vector space with a scalar product is called an inner product space. The most
common example of an inner produce space is the plane (R2 , +, ·) with the usual dot
product: given u(u1 , u2 ), v(v1 , v2 ) ∈ R2 then

hu, vi = u1 v1 + u2 v2 .

Remember the vector space of continuous functions (C[0, 1], +, ·)? It turns out that this
space also has an inner product. Given two functions f, g ∈ C[0, 1] we define the scalar
product as follows: Z 1
hf, gi = f (x)g(x)dx.
0
Let us verify that this is indeed a scalar product. It is clearly symmetrical as we have
Z 1 Z 1
hf, gi = f (x)g(x)dx = g(x)f (x)dx = hg, f i.
0 0

It also satisfies the linearity conditions as we have

Z 1 Z 1 Z 1
hf + h, gi = (f (x) + h(x))g(x)dx + f (x)g(x)dx + h(x)g(x)dx = hf, gi + hh, gi
0 0 0
Z 1 Z 1
hr · f, gi = rf (x)g(x)dx = r f (x)g(x)dx = rhf, gi.
0 0
Lastly we verify the positivity condition. As the square of any function is non-negative
and the integral of any non-negative function is a non-negative number we can write:
Z 1
hf, f i = f 2 (x)dx ≥ 0.
0

Additionally, if hf, f i = 0 then the area under the graph of the non-negative function f 2
is zero. This can only happen if f 2 is the zero function which means that f = 0, with
this fully verifying the positivity condition. We conclude that hf, gi as defined above is
indeed a scalar product.
The most important result about inner products is the Cauchy-(Bunyakovsky)-Schwarz
inequality. As we will see, with the help of this inequality we will be able to define angles
between vectors.

5
Theorem 3 (Cauchy-Schwarz). Suppose (V, +, ·, h·, ·i) is an inner product space. Then
for any u, v ∈ V we have
hu, vi2 ≤ hu, uihv, vi.

Proof. The proof is very slick application of the quadratic formula. We fix u, v ∈ V .
Suppose t ∈ R is any real number. By the positivity of scalar products we have

hu + tv, u + tvi ≥ 0.

A calculation using the linearity of the inner product gives

hu, ui + 2thu, vi + t2 hv, vi = hu, ui + 2hu, tvi + htv, tvi = hu + tv, u + tvi ≥ 0.

According to this estimate, the parabola f (t) = hv, vit2 + 2hu, vit + hu, ui stays above
the oX axis. This means that the following quadratic equation (in t) has at most one
solution:
hv, vit2 + 2hu, vit + hu, ui = 0.
This implies that the discriminant ∆ of this equation has to be non-positive:

∆ = (2hu, vi)2 − 4hv, vihu, ui ≤ 0.

Dividing this estimate with 4 we arrive at the desired inequality:

hu, vi2 − hv, vihu, ui ≤ 0.

Remark 2. In particular case of the plane, the Cauchy-Schwarz inequality says that for
u(u1 , u2 ), v(v1 , v2 ) ∈ R2 we have

(u1 v1 + u2 v2 )2 = hu, vi2 ≤ hu, uihv, vi = (u21 + u22 )(v12 + v22 ).

In case of the vector space of continuous functions C[0, 1], the Cauchy-Schwarz inequality
says that for f, g ∈ C[0, 1] we have
Z 1 Z 1 Z 1
2
f (x)g(x)dx ≤ f (x)dx g 2 (x)dx.
0 0 0

With the notion of inner product under our belt, we can introduce angles between
vectors. Given two vectors u, v we would like to understand what is the angle made by
these vectors with vertex at the zero vector 0.

6
 Even though it is hard to visualize angles in abstract vector spaces, we can compute
the cosine of the angle u0v
d using the following formula:

d = p hu,p
cos(u0v)
vi
. (3)
hu, ui hv, vi

This formula is well known for the standard dot product of the plane (R2 , +, ·), and of
course this is where the analogy comes from in the general case.
By taking arccos of both sides we end up with the following formula for the actual
measure of the angle u0v.
d
 hu, vi 
m(u0v) = arccos
d p p .
hu, ui hv, vi
As a last observation, notice that the Cauchy-Schwarz inequality proved above is needed
for the right hand side to make sense. Indeed, the arccos function is defined on the
interval [−1, 1] and takes its values in [0, π]. Hence we need something that assures that
hu, vi
−1 ≤ p p ≤1
hu, ui hv, vi
for any vectors u, v so that we can apply arccos to this expression. Clearly, this condition
is just the conclusion of the Cauchy-Schwarz inequality.
As in the plane, one can also measure angles with vertex not necessary at the 0
vector. Suppose u, v, w ∈ V , then for the angle uvw
d we have the following formulas:

hu − v, w − vi
cos(uvw)
d =p p ,
hu − v, u − vi hw − v, w − vi
 
hu − v, w − vi
m(uvw)
d = arccos p p .
hu − v, u − vi hw − v, w − vi
Problem 2. Show that the continuous functions f (x) = x − 21 , g(x) = 1 are perpendic-
ular.
0g) = 90o . This is equivalent with showing that
Solution. We have to show that m(fd

cos(fd
0g) = 0.

7
 0g) in (3), we have to show that hf, gi = 0. So we
Looking at the definition of a cos(fd
compute this last scalar product:
Z 1 Z 1
1 1 x2 1 x 1 1 1
hf, gi = (x − )1dx = x − dx = − = − = 0.
0 2 0 2 2 0 2 0 2 2

Problem 3. Given vectors a, b, c ∈ R2 , show using vector geometry that the altitudes in
the triangle abc∆ are concurrent.

Proof. Let aa0 and bb0 be altitudes according to the figure. Let o be the intersection of
aa0 with bb0 . To prove that o is on the altitude stemming from the vertex c, we have to
show that the vector c − o is perpendicular to the vector b − a. We show this with the
help of the scalar product.
From our conditions it follows that bo is perpendicular to ca and ao is perpendicular
to cb so we can write
0 = hb − o, c − ai,
0 = ha − o, b − ci.
By using the linearity of the scalar product in the first identity we can write:
0 =hb − o, c − ai = hb, c − ai − ho, c − ai
=hb, ci − hb, ai − ho, ci + ho, ai.
Similarly, the second identity becomes
0 = ha, bi − ha, ci − ho, bi + ho, ci.
After we add these last two identities, a few terms cancel and we end up with
0 = hb, ci + ho, ai − ha, ci − ho, bi.
Observe that the right hand side is just equal to hb − a, c − oi. Hence we have
0 = hb − a, c − oi,
which is what we wanted to prove.

8
 We saw that using inner products one can measure angles. It turns out that one
can measure distances as well. But what is distance anyway? We try to capture this
important notion with the next definition:

Definition 5. Given a vector space (V, +, ·), a function d : V × V → R is a distance if

for every u, v, w ∈ V the following conditions hold:

• (symmetry) d(u, v) = d(v, u),

• (triangle inequality) d(u, w) ≤ d(u, v) + d(v, w),

• (positivity) d(u, v) ≥ 0 and if d(u, v) = 0 then u = v.

The following figure should make it clear why the second condition is called the
triangle inequality.

As advertised in the beginning, the following theorem says that given an inner product
on a vector space, there is a natural way to introduce a distance:

Theorem 4. Suppose (V, +, ·, h·, ·i) is an inner product space. Given u, v ∈ V the
expression p
d(u, v) = hu − v, u − vi
defines a distance on V .

Proof. The proof is just a verification of the conditions in definition of distance. The
symmetry condition is the easiest:
p p p
d(u, v) = hu − v, u − vi = (−1)2 hv − u, v − ui = hv − u, v − ui = d(v, u).

Next we verify the triangle inequality. Using the linearity of the inner product, we can
start writing:
p p
d(u, v) = hu − v, u − vi = h(u − w) + (w − v), (u − w) + (w − v)i
p
= hu − w, (u − w) + (w − v)i + hw − u, (u − w) + (w − v)i
p
= hu − w, u − wi + 2hu − w, w − vi + hw − v, w − vi,

Now the Cauchy-Schwarz inequality implies that

p
hu − v, w − vi ≤ hu − w, u − wihw − v, w − vi.

9
Hence, we can majorize the right hand side above the following way:
q p
d(u, v) ≤ hu − w, u − wi + 2 hu − w, u − wihw − v, w − vi + hw − v, w − vi.

Observe that under the square root we have a perfect square, hence we can continue:
qp p
d(u, v) ≤ ( hu − w, u − wi + hw − v, w − vi)2
p p
= hu − w, u − wi + hw − v, w − vi
= d(u, w) + d(v, w),

which is the triangle inequality. Finally we verify the positivity condition. By the
positivity property of inner products we obtain that hu − v, u − vi ≥ 0 hence also
p
d(u, v) = hu − v, u − vi ≥ 0.

If d(u, v) = 0 then also hu − v, u − vi = 0, which by the positivity property of the

inner product implies that u − v = 0. This verifies the last condition (positivity) in the
definition of distance.

Remark 3. In case of (R2 , +, ·, h·, ·i) the distance given by the previous theorem is just
the well known Euclidean distance. Indeed, for u(u1 , u2 ), v(v1 , v2 ) ∈ R2 we have
p p
d(u, v) = hu − v, u − vi = (u1 − v1 )2 + (u2 − v2 )2 .

In case of the ”weird” example (C[0, 1], +, ·, h·, ·i) the distance is
s
p Z 1
d(f, g) = hf − g, f − gi = (f (x) − g(x))2 dx.
0

Problem 4. Suppose f, g ∈ C[0, 1] defined by f (x) = x + 1 and g(x) = x2 + 1. Compute

the distance d(f, g).
p
Solution. By the above formula we have d(f, g) = hf − g, f − gi. Now we compute
this last expression:
s s
p Z 1 Z 1
hf − g, f − gi = 2 2
(x − x ) dx = (x2 + x4 − 2x3 )dx
0 0
r r
1 1 1 10 + 6 − 15
= + − = .
3 5 2 30
q
1
Consequently, d(f, g) = 30
.

10