82 Chapter 2

3(M - 6)
■x, + -x, + [x3 + (M - 10/3)d,]
4(M-10/3)' (M - 10/3)

or

+ -X-.i + 2d,,
X = —Xi
4 ' 4 '
which is of the required form of Equation (2.3). Note that this representation is
not unique. Another representation for x is

l ] Ί
- Λ
X = —X7 H Xi H di.
l J
2 2 3 '
Note from Figure 2.18 that n = 2, k = 3, and I = 2 in this problem. Hence,
min{(« + 1), (k + i )} = 3, and hence, both representations use the upper bound
number of terms prescribed by Corollary 2.1. In this case, a lower bound on the
number of terms required in any representation of x = (4, 3) also happens to be
three.
Some Insights into the Equality Constrained Polyhedron
Consider the nonempty polyhedral set X = {x : Ax = b, x > 0} where A is m x n
and rank(A) = m. Note that by equivalently replacing each equality constraint by
two inequalities, all the discussion in this section continues to hold true for this
case as well. In particular, consider the extreme points of X. As before, these are
points in X formed by the intersection of n linearly independent hyperplanes.
However, from the equality constraints, we know that m linearly independent
hyperplanes are always binding at any feasible solution. Hence, at any extreme
point x of X, there must be some (n - tri) additional hyperplanes binding from
the nonnegativity constraints, say, xN = 0, which together with Ax = b pro-
duce n linearly independent equations in n unknowns, thereby yielding x as the
unique solution. Similarly, one can characterize edges and adjacent extreme
points of X following Section 2.6, and note as before that extreme directions of
X correspond to extreme points of
D = {d : Ad = 0,ld = l,d > 0}.
Finally, in light of the remark concerning extreme points of X, the reader may
find it instructive to read through the proof of Theorem 2.1, thinking of Xas the
equality constrained set.
EXERCISES
[2.1] Show that the vectors
ίίλ
a, = 0 , 1 and *3 -
vh
Linear Algebra, Convex Analysis, and Polyhedral Sets 83

form a basis for R . Supposing that a 2 is replaced by indicate whether the

new set of vector still forms a basis of R .

"1 ~l

[2.2] Which of the following collection of vectors form a basis of R , span R ,

or neither?

a. aj = 2
rn a2 =
f-n0 a
foi
0
3 =
lu lu
?

1-iJ
(l) f2i
b. aj = 3 a2 = 0

f-1 \ (ολ fU ^_3Λ

c. a! = 2 a2 = 1 a
3 = 2 2
l°J
ΐ

I 3/ l3j V 4y
rn f-3 \ ί -18
λ
d. aj = 2 a 2 = -1 a3 =
lu
5

I 2/ I 7/
rn f-1 \ ί°Ί
e. aj
w
= 4 a 2 = -4
l-i /
? a3 = 1
l°J
[2.3] Let
/ 5/2Ì
f-l>
2 a2 =
i3l 3
2 , a3
a
l = 5

l°v 15J V5J

Are these vectors linearly independent? Do they span R ?

[2.4] Let a1,a2,...,a/t form a basis for R". Show that a],a 2 ,...,a^. are linearly
independent. Also show that k = n.
[2.5] Does the following matrix have an inverse? If the answer is yes, find A~ :
"1 - 4 4 0"
2 5 0 1
0 2 0 1
1 3 2 2_
[2.6] Find the inverse of the following triangular matrix:
"2 4 - 3 -Λ
0 4 3 2
0 0 2 5
0 0 0-1
84 Chapter 2

[2.7] Let A be a 3 x 5 matrix. Consider the matrix C obtained by weighting the

columns of A with scalars 1, 3/2, - 2 , 4, and - 1 , respectively, and adding them
up. Write C as a matrix product. Similarly, let D be the weighted sum of the
rows of A, using weights 2, - 2 , and 3/2, respectively. Write D as a matrix
product. Likewise, let E be a matrix whose first row is the sum of the rows of A,
whose second row is the third row of A, and whose third row is obtained by
summing: thrice the first row of A, -5/2 times the second row of A, and twice
the third row of A. Write E as a matrix product.
[2.8] Let B be an invertible matrix. Show that B" is unique.
[2.9] Suppose that &x,a2,...,&n form a basis of/?" and y = Ajaj +À2&2 + ·■· +
/L,a„ with Àj = 0. Prove that aj,...,a,-_j,y, a , +1 ,...,a„ do not form a basis of R".
[2.10] Let
"B 0"
T I

where B is an m χ m invertible matrix, I is a k x k identity matrix, 0 is an m x k

zero matrix, and T is an arbitrary k x m matrix. Show that A has an inverse and
that

B- 1 0
A'1 =
TB1 I

[2.11] Show that if A and B are n χ η matrices that are both invertible, then
(AB) -1 = B _1 A _1 .

[2.12] Let A be an » x « invertible matrix. Show that A' has an inverse and that
(A')" 1 = (A" 1 )'.
[2.13] Let B be an invertible matrix with nonnegative entries. Show that every
row of B~ has a least one positive entry.
[2.14] Let A = (a 1 ,a 2 ,...,a,-,...,a / „) be an invertible m χ m matrix. Show that
A ' a,· = e ,· where e ■
, is a vector of zeros except for a 1 at positiony.
[2.15] Let A be an « x « matrix. Suppose that B is an n χ n matrix such that AB
= I. Is it necessarily true that A has an inverse? Is it necessarily true that B =
A"1?
[2.16] If the rth row of a square nonsingular matrix B is multiplied by a scalar
λ * 0, what changes would result in B~ ?
[2.17] If the rth column of a square nonsingular matrix B is multiplied by a
scalar λ Φ 0, what changes would result in B~ ?
Linear Algebra, Convex Analysis, and Polyhedral Sets 85

[2.18] Find the rank of the following matrices:

"1 0 1 - f
(i) A = 2 3 4 1
10 5-3
"-2 1 -3"
(ii) A = 2 4-2
4 3 1
[2.19] Find the determinants of the following matrices.
"10 f
a. A 2 2-1
0 1
1 0 -2
- 3 1 -1
b. A =
-2 2 -1
1 2 2
'2 -2 ί
c. 1 3 -2
3 -2 - 3
[2.20] Solve the following system by Cramer's Rule:
2x, x
2
5x, + 2xi
[2.21] Demonstrate by enumeration that every basis matrix of the following
system is triangular:
Χγ + X-J + X4 — 4
~X\ "H X"2 ~^~ X$ =
2
— X2 ~ X4. =
- 3
- x-i = -2.

tig system of equations:

Xj + 2x2 + X3 = 1
x
l + *2 -
*3 = 3
X] + 3x2 + X3 = -4

Without resolving the system, what is the solution if the right-hand-side of the
first equation is changed from 1 to 2?
[2.23] Show that the determinant of a square triangular matrix is the product of
the diagonal entries.
[2.24] Find all basic solutions of the following system:
86 Chapter 2

-X] + 2x2 + X3 + 3x4 - 2x5 = 4

xx - 2x2 + 2x4 + x5 = 2.
[2.25] Determine whether the following system possesses: (a) no solution, (b) a
unique solution, or (c) many (how many?) solutions.
Xi + 3x2 + X3 — ΧΛ — 1
5x2 _ 6x3 + x 4 = 0
X] - 2x2 + 4x3 = 2.

[2.26] Construct a general solution of the system Ax = b where A is an m χ η

matrix with rank m. What is the general solution of the following system?
Xi — 2x2 + X3 = 2
-Xj + 3x2 + 2x3 = 6.

[2.27] Consider the system Ax = b where A = [a|,a 2 ,...,a„] is an m χ η matrix

of rank m. Let x be any solution of this system. Starting with x, construct a basic
solution. (Hint: Suppose that xx,...,xp * 0 and xp+x,...,xn = 0. If p > m, repre-
sent one of the columns a , fory = 1, ..., p, as a linear combination of the
remaining vectors. This results in a new solution having a smaller number of
nonzero variables. Repeat the process.)
[2.28] Which of the following sets are convex and which are not?
a. {(xx,x2) : X\ + x2 > 3}.
b. {(xj,x2,X3) : xj + 2x2 < 1, xj - 2x3 < 2}.
c. {(xj,x2) : x 2 - 3x] = 0}.
d. {(x],x 2 ,x 3 ) : x 2 > Xj , X] + 2x2 + X3 < 4}.
e. {(xi,x 2 ) : xi = 3 , |x 2 | < 4}.
f. {(x 1; x 2 ,x 3 ) : x3 = |x 2 |, X! < 3}.
[2.29] Which of the following functions are convex, concave, or neither?
a. f(x) = x 2 .
b. / ( x l 5 x 2 ) = e"*1"*2 + 2x,2 - 2xx.
c. f(xx, x 2 ) = maximum{fx (xx,x2), f2 (xx,x2)} where / , (^, x2 ) = 3x2
"7 2
+ x2 a n d f2Ìxl>x2) = ^xl _ 5*2-
where and
d- /(Xi,X2) = minimum { ^ ( x ^ ) , f2(x\*x2)} Ai') /2O
are defined in Part (c).
2 2 2
= —x x
e. y (Xi,Χ7 >-*3) \ ~~ ^ X 2 ~ 3 ~^~ 2XjX 2 ~~ %2X3 ~^~ 1 ~^~ 5x-\.
—
1. J (Xi , X 2 ) — Xi + X 2 2X]X 2 + X 2 .
Linear Algebra, Convex Analysis, and Polyhedral Sets 87

[2.30] Consider the set {(x l5 x 2 ) : ~x\ + x2 - 2 , x\ + 2x2 -8> x\ - 0> x2 - 0 } ·

What is the minimum distance from (-3,4) to the set? What is the point in the
set closest to (-3,4)? Repeat with the point (6,4).

[2.31] Let a i = (^), a 2 = (*), ^ = ( " i j , « 4 = |_4), and a 5 = ( ^ ) . Illus-

trate geometrically the collection of all convex combinations of these five

points.
[2.32] Show that a hyperplane H = {x: px = k) and a half-space H+ = {x: px >
k] are convex sets.
[2.33] Show that/is convex if and only if its epigraph ={(x,y) : x e R", y e
R, y > f(x)} is a convex set. Similarly, show that/is concave if and only if its
hypograph = {(x,y) : x e R", y e R , y < /(x)} is a convex set.
[2.34] Show that a differentiable function / is convex if and only if the
following inequality holds for each fixed point x 0 in Rn: f(x) > /(XQ) +
V/(x 0 )'(x - XQ) for all x e R", where V/(x 0 ) is the gradient vector off at x 0
given by

> ( X Q ) df(x0) g/(x 0 )Y

cbcj ' 9x2 ' ' δχη J
[2.35] Show that the set of feasible solutions to the following linear program
forms a convex set:
Minimize ex
subject to Ax = b
x>0.
[2.36] Consider the set X = {{x\,xj) : x\ + x2 ^ 2, x2 < 4, x l5 x 2 > 0}.
Find a hyperplane / / such that X and the point (1,-2) are on different sides of
the hyperplane. Write the equation of the hyperplane.
[2.37] If 5 is an open set, show that the problem
Maximize ex
subject to x e S
where c Φ 0 possesses no optimal point. {Note: S is open if for each x 0 e S,
there is an ε > 0 such that ||x - x 0 | < ε implies that x e S.)
[2.38] Show that C is a convex cone if and only if x and y e C imply that
λχ + //y e C for all λ > 0 and μ > 0.
[2.39] Show that if C is a convex cone, then C has at most one extreme point,
namely, the origin.
88 Chapter 2

[2.40] Find the extreme points of the region defined by the following
inequalities:
Xj + 2x2 + x3 < 5
—X\ + x2 + 2x3 < 6
Xi , Xo, X3 ^ 0.

(///«/: Consider n = 3 intersecting defining hyperplanes at a time.)

[2.41] Why does the following set have directions? Find all its extreme
directions:
-Xj + 2x2 = 3
2x] - 2x2 - X3 < 2
x3 > 1
Xi, Xj , X3 — 0.

(Hint: Enumerate the extreme points of its normalized set of directions. What
else needs to be checked?)
[2.42] Find all extreme points of the following polyhedral set:
X= {(xi,x 2 ,x 3 ) : X\ - x2 + Χτ, ^ Ι,Λ^Ι - 2x2 < 4,xl5x2,X3 > 0}.
Does X have any recession directions? Why?
[2.43] Let X= {x : Ax < b} c R" and let x 0 e X be such that fewer than n
linearly independent hyperplanes defining X are active at x 0 . Show that x 0
cannot be an extreme point of X.
[2.44] Prove in detail that a polyhedral set X is bounded if and only if it has no
directions.
[2.45] Consider the nonempty polyhedral set X = {x : Ax = b, x > 0}. Show
directly by definition that d is a direction of the set if and only ifd * 0, Ad = 0,
andd > 0. Obtain analogous results if Ax = b is replaced byAjx = bj,
A2x > b 2 .
[2.46] Given a nonempty polyhedron X, a face F of X is also defined as the
intersection of a supporting hyperplane of X with X itself. (A hyperplane
supports Χ'ιΐX lies completely in one half-space defined by the hyperplane and
the intersection of the hyperplane with X is nonempty.) Show the equivalence
between this definition and the one given in Section 2.6.
[2.47] You are given the following polyhedral set. Identify the faces, extreme
points, extreme directions, and extreme rays of the set.
Xj - x2 + x3 < 10
2χλ - x2 + 2x3 < 40
3xj - 2x2 + 3x3 < 50
Xi, X2, X3 ^ 0.
Linear Algebra, Convex Analysis, and Polyhedral Sets 89

[2.48] Let X ={(xx,x2) : χλ - x2 < 3, -x1 + 3x2 < 3, xx > -3}. Find all
extreme points of X and represent x = (0, 1) as a convex combination of the
extreme points.
[2.49] Answer the following questions and provide a brief explanation or
illustration:
a. Is it possible for X in Equation (2.1) to be empty but D in Equation
(2.2) to be nonempty?
b. Is there a relationship between redundancy and degeneracy of a
polyhedral set?
c. Does degeneracy imply redundancy in two dimensions?
d. If the intersection of a finite number of half-spaces is nonempty,
then this set has at least one extreme point. True or false? Explain.
e. An unbounded «-dimensional polyhedral set can have at most n
extreme directions. True or false? Explain.
f. What is the maximum (actual) dimension of X = {x : Ax = b, x
> 0}, where A is m x n of rank r, with r<m<rf!
[2.50] Find all extreme points and extreme directions of the following poly-
hedral set:
X = \(X\ ,X2 5-^-3 , %4 ) · —X\ "t" %2 ~~ 3 '
—2X] — Xy + 2%4 < 2, X],X2,X2,XQ ^ 0}.

Represent x = (1, 1, 1, 2) as a convex combination of the extreme points of X

plus a nonnegative combination of the extreme directions of X.
[2.51] Show that an unbounded polyhedral set of the form {x : Ax = b, x > 0 }
has at least one extreme direction. In particular, using the normalized direction
set D in lieu of Xin the proof of Theorem 2.1, show how you can start with any
direction d in D and reduce it to an extreme direction.
[2.52] Consider the polyhedral set X = {x : px = k] where p is a nonzero
vector and A: is a scalar. Show that X has neither extreme points nor extreme
rays. How do you explain this in terms of the general Representation Theorem?
[2.53] LetX= {x : Ax = b,x > 0} where A is an w x « matrix with rank m.
Show that d is an extreme direction of X if and only if d is a positive multiple of
the vector (-yy,0,0,...,l,0,...,0)' where the 1 appears in positionj, and where:

y7 = ΒΛ7<Ο,
A = [B, N] where B is an m x m invertible matrix
a = a column of N.

(Hint: Consider extreme points of the normalized direction set.) Illustrate by the
following system:
90 Chapter 2

X] - 3X2 + *3 = 0
-3x) + X2 + x4 = 3
Xi, Xo, X3 , X4 ^ U.

[2.54] In the proof of Theorem 2.1, consider the direction d given by

(x v - x,(v))/(9v. Show how you would algebraically obtain d, given the
extreme point xv of X, v> k.
[2.55] Consider the polyhedral set X = {x : Qx < q}, where Q is m χ η.
Mathematically characterize the statement that an entire line belongs to X. What
does this imply about the rank of Q? Show that a nonempty polyhedron has
extreme points if and only if it contains no lines.
[2.56] Based on Theorem 2.1 and its proof, construct a polynomial-time
algorithm of complexity bounded by 0{n ) in order to obtain a representation
of x e X in terms of no more than (n + 1) extreme points and n extreme
directions of X. Show that the number of extreme points needed in the
representation by this algorithm is no more than the bound provided by
Corollary 2.1 when X is a bounded set.
NOTES AND REFERENCES
1. Sections 2.1 through 2.3 present a quick review of some relevant results
of vector and matrix algebra.
2. Sections 2.4 and 2.5 give some basic definitions and properties of convex
sets, convex cones, and convex functions. For more details the reader
may refer to Eggleston [1958], Mangasarian [1969], Bazaraa, Sherali, and
Shetty [2006], and Rockafellar [1970].
3. Correspondence between bases and extreme points is established in the
next chapter in Section 3.2, and an algebraic characterization of extreme
directions is presented in Exercise 2.53.
4. The representation theorem for polyhedral sets evolved from the work of
Minkowski [1910] and Goldman and Tucker [1956]. The result is also
true for (nonpolyhedral) convex sets that contain no lines. See
Rockafellar [1970] and Bazaraa and Shetty [1976]. The geometric proof
of Theorem 2.1 in Section 2.6 is taken from Sherali [1987b]. (Parts of the
proof, parts of Example 2.7, and Figure 2.17 are reprinted from H. D.
Sherali, "A Constructive Proof of the Representation Theorem for
Polyhedral Sets Based on Fundamental Definitions," American Journal
of Mathematical and Management Sciences, Vol. 7 (1987), 253-270,
Copyright © 1987 by the American Sciences Press, Inc., 20 Cross Road,
Syracuse, New York 13224. Reprinted by permission.) For further
geometric insights, see Akgul [1988], Grunbaum [1967], Murty [1983,
1985], and Sommerville [1958].