EE364a, Winter 2012-13 Prof. S.

Boyd
EE364a Homework 1 solutions
2.9 Voronoi sets and polyhedral decomposition. Let x
0
, . . . , x
K
R
n
. Consider the set of
points that are closer (in Euclidean norm) to x
0
than the other x
i
, i.e.,
V = x R
n
[ |x x
0
|
2
|x x
i
|
2
, i = 1, . . . , K.
V is called the Voronoi region around x
0
with respect to x
1
, . . . , x
K
.
(a) Show that V is a polyhedron. Express V in the form V = x [ Ax _ b.
(b) Conversely, given a polyhedron P with nonempty interior, show how to nd
x
0
, . . . , x
K
so that the polyhedron is the Voronoi region of x
0
with respect to
x
1
, . . . , x
K
.
(c) We can also consider the sets
V
k
= x R
n
[ |x x
k
|
2
|x x
i
|
2
, i ,= k.
The set V
k
consists of points in R
n
for which the closest point in the set x
0
, . . . , x
K

is x
k
.
The sets V
0
, . . . , V
K
give a polyhedral decomposition of R
n
. More precisely, the
sets V
k
are polyhedra,

K
k=0
V
k
= R
n
, and int V
i
int V
j
= for i ,= j, i.e., V
i
and
V
j
intersect at most along a boundary.
Suppose that P
1
, . . . , P
m
are polyhedra such that

m
i=1
P
i
= R
n
, and int P
i

int P
j
= for i ,= j. Can this polyhedral decomposition of R
n
be described as
the Voronoi regions generated by an appropriate set of points?
Solution.
(a) x is closer to x
0
than to x
i
if and only if
|x x
0
|
2
|x x
i
|
2
(x x
0
)
T
(x x
0
) (x x
i
)
T
(x x
i
)
x
T
x 2x
T
0
x +x
T
0
x
0
x
T
x 2x
T
i
x +x
T
i
x
i
2(x
i
x
0
)
T
x x
T
i
x
i
x
T
0
x
0
,
which denes a halfspace. We can express V as V = x [ Ax _ b with
A = 2

(x
1
x
0
)
T
(x
2
x
0
)
T
.
.
.
(x
K
x
0
)
T

, b =

x
T
1
x
1
x
T
0
x
0
x
T
2
x
2
x
T
0
x
0
.
.
.
x
T
K
x
K
x
T
0
x
0

.
1
(b) Conversely, suppose V = x [ Ax _ b with A R
Kn
and b R
K
. We can
pick any x
0
x [ Ax b, and then construct K points x
i
by taking the mirror
image of x
0
with respect to the hyperplanes x [ a
T
i
x = b
i
. In other words, we
choose x
i
of the form x
i
= x
0
+ a
i
, where is chosen in such a way that the
distance of x
i
to the hyperplane dened by a
T
i
x = b
i
is equal to the distance of
x
0
to the hyperplane:
b
i
a
T
i
x
0
= a
T
i
x
i
b
i
.
Solving for , we obtain = 2(b
i
a
T
i
x
0
)/|a
i
|
2
2
, and
x
i
= x
0
+
2(b
i
a
T
i
x
0
)
|a
i
|
2
a
i
.
(c) A polyhedral decomposition of R
n
can not always be described as Voronoi regions
generated by a set of points x
1
, . . . , x
m
. The gure shows a counterexample in
R
2
.
P
1
P
2
P
3
P
4

P
1

P
2
H
1
H
2
R
2
is decomposed into 4 polyhedra P
1
, . . . , P
4
by 2 hyperplanes H
1
, H
2
. Suppose
we arbitrarily pick x
1
P
1
and x
2
P
2
. x
3
P
3
must be the mirror image of x
1
and x
2
with respect to H
2
and H
1
, respectively. However, the mirror image of x
1
with respect to H
2
lies in

P
1
, and the mirror image of x
2
with respect to H
1
lies
in

P
2
, so it is impossible to nd such an x
3
.
2.12 Which of the following sets are convex?
(a) A slab, i.e., a set of the form x R
n
[ a
T
x .
(b) A rectangle, i.e., a set of the form x R
n
[
i
x
i

i
, i = 1, . . . , n. A
rectangle is sometimes called a hyperrectangle when n > 2.
(c) A wedge, i.e., x R
n
[ a
T
1
x b
1
, a
T
2
x b
2
.
2
(d) The set of points closer to a given point than a given set, i.e.,
x [ |x x
0
|
2
|x y|
2
for all y S
where S R
n
.
(e) The set of points closer to one set than another, i.e.,
x [ dist(x, S) dist(x, T),
where S, T R
n
, and
dist(x, S) = inf|x z|
2
[ z S.
(f) The set x [ x +S
2
S
1
, where S
1
, S
2
R
n
with S
1
convex.
(g) The set of points whose distance to a does not exceed a xed fraction of the
distance to b, i.e., the set x [ |x a|
2
|x b|
2
. You can assume a ,= b and
0 1.
Solution.
(a) A slab is an intersection of two halfspaces, hence it is a convex set (and a poly-
hedron).
(b) As in part (a), a rectangle is a convex set and a polyhedron because it is a nite
intersection of halfspaces.
(c) A wedge is an intersection of two halfspaces, so it is convex set. It is also a
polyhedron. It is a cone if b
1
= 0 and b
2
= 0.
(d) This set is convex because it can be expressed as

yS
x [ |x x
0
|
2
|x y|
2
,
i.e., an intersection of halfspaces. (For xed y, the set
x [ |x x
0
|
2
|x y|
2

is a halfspace; see exercise 2.9).

yS
2
x [ x +y S
1
=

yS
2
(S
1
y),
the intersection of convex sets S
1
y.
(g) The set is convex, in fact a ball.
x [ |x a|
2
|x b|
2

= x [ |x a|
2
2

2
|x b|
2
2

= x [ (1
2
)x
T
x 2(a
2
b)
T
x + (a
T
a
2
b
T
b) 0
If = 1, this is a halfspace. If < 1, it is a ball
x [ (x x
0
)
T
(x x
0
) R
2
,
with center x
0
and radius R given by
x
0
=
a
2
b
1
2
, R =

2
|b|
2
2
|a|
2
2
1
2
+|x
0
|
2
2

1/2
.
2.15 Some sets of probability distributions. Let x be a real-valued random variable with
prob(x = a
i
) = p
i
, i = 1, . . . , n, where a
1
< a
2
< < a
n
. Of course p R
n
lies in
the standard probability simplex P = p [ 1
T
p = 1, p _ 0. Which of the following
conditions are convex in p? (That is, for which of the following conditions is the set of
p P that satisfy the condition convex?)
(a) Ef(x) , where Ef(x) is the expected value of f(x), i.e., Ef(x) =

n
i=1
p
i
f(a
i
). (The function f : R R is given.)
(b) prob(x > ) .
(c) E[x
3
[ E[x[.
(d) Ex
2
.
(e) Ex
2
.
(f) var(x) , where var(x) = E(x Ex)
2
is the variance of x.
(g) var(x) .
(h) quartile(x) , where quartile(x) = inf [ prob(x ) 0.25.
(i) quartile(x) .
Solution. We rst note that the constraints p
i
0, i = 1, . . . , n, dene halfspaces,
and

n
i=1
p
i
= 1 denes a hyperplane, so P is a polyhedron.
The rst ve constraints are, in fact, linear inequalities in the probabilities p
i
.
4
(a) Ef(x) =

n
i=1
p
i
f(a
i
), so the constraint is equivalent to two linear inequalities

n

i=1
p
i
f(a
i
) .
(b) prob(x ) =

i: a
i

p
i
, so the constraint is equivalent to a linear inequality

i: a
i

p
i
.
(c) The constraint is equivalent to a linear inequality
n

i=1
p
i
([a
3
i
[ [a
i
[) 0.
(d) The constraint is equivalent to a linear inequality
n

i=1
p
i
a
2
i
.
(e) The constraint is equivalent to a linear inequality
n

i=1
p
i
a
2
i
.
The rst ve constraints therefore dene convex sets.
(f) The constraint
var(x) = Ex
2
(Ex)
2
=
n

i=1
p
i
a
2
i
(
n

i=1
p
i
a
i
)
2

is not convex in general. As a counterexample, we can take n = 2, a
1
= 0, a
2
= 1,
and = 1/5. p = (1, 0) and p = (0, 1) are two points that satisfy var(x) ,
but the convex combination p = (1/2, 1/2) does not.
(g) This constraint is equivalent to
n

i=1
p
i
a
2
i
(
n

i=1
p
i
a
i
)
2
= b
T
p p
T
Ap ,
where b
i
= a
2
i
and A = aa
T
. We write this as
p
T
Ap b
T
p + 0.
This denes a convex set, since the matrix aa
T
is positive semidenite.
5
To show this set is convex for A S
+
, we rst make the observation that
x
T
Ay
x
T
Ax +y
T
Ay
2
which follows from (x y)
T
A(x y) 0. Now choose x, y ( and show z =
x + (1 )y (:
z
T
Az +b
T
z +
=
2
x
T
Ax + (1 )
2
y
T
Ay + 2(1 )x
T
Ay +b
T
(x + (1 )y) +

2
x
T
Ax + (1 )
2
y
T
Ay +(1 )(x
T
Ax +y
T
Ay) +b
T
(x + (1 )y) +
= x
T
Ax +b
T
x + (1 )y
T
Ay + (1 )b
T
y + + (1 )
= (x
T
Ax +b
T
x +) + (1 )(y
T
Ay +b
T
y +)
0
Therefore z ( and the set is convex.
Let us denote quartile(x) = f(p) to emphasize it is a function of p. The gure
illustrates the denition. It shows the cumulative distribution for a distribution p with
f(p) = a
2
.

prob(x )
a
1
a
2
a
n
p
1
p
1
+p
2
p
1
+p
2
+ +p
n1
0.25
1
(h) The constraint f(p) is equivalent to
prob(x ) < 0.25 for all < .
If a
1
, this is always true. Otherwise, dene k = maxi [ a
i
< . This is a
xed integer, independent of p. The constraint f(p) holds if and only if
prob(x a
k
) =
k

i=1
p
i
< 0.25.
This is a strict linear inequality in p, which denes an open halfspace.
6
(i) The constraint f(p) is equivalent to
prob(x ) 0.25 for all .
Here, let us dene k = maxi [ a
i
. Again, this is a xed integer, independent
of p. The constraint f(p) holds if and only if
prob(x a
k
) =
k

i=1
p
i
0.25.
If < a
1
, then no p satises f(p) , which means that the set is empty. Thus,
the constraint f(p) is a linear inequality on p.
2.23 Give an example of two closed convex sets that are disjoint but cannot be strictly
separated.
Solution. Take C = x R
2
[ x
2
0 and D = x R
2
+
[ x
1
x
2
1.
2.28 Positive semidenite cone for n = 1, 2, 3. Give an explicit description of the positive
semidenite cone S
n
+
, in terms of the matrix coecients and ordinary inequalities, for
n = 1, 2, 3. To describe a general element of S
n
, for n = 1, 2, 3, use the notation
x
1
,

x
1
x
2
x
2
x
3

x
1
x
2
x
3
x
2
x
4
x
5
x
3
x
5
x
6

.
Solution. A symmetric matrix X is positive semidenite if and only if all principal
minors (determinants of symmetric submatrices) are nonnegative. For n = 1 the
condition is just x
1
0. For n = 2 the condition is
x
1
0, x
3
0, x
1
x
3
x
2
2
0.
For n = 3 the condition is
x
1
0, x
4
0, x
6
0, x
1
x
4
x
2
2
0, x
4
x
6
x
2
5
0, x
1
x
6
x
2
3
0
and
x
1
x
4
x
6
+ 2x
2
x
3
x
5
x
1
x
2
5
x
6
x
2
2
x
4
x
2
3
0.
7