The Geometry of Linear Programming*

Matthew Hough
March 20, 2023

Contents
1 Polyhedra 2

2 Extreme points, vertices, and basic feasible solutions 2

3 Polyhedra in standard form 5

4 Degeneracy 8

5 Existence of extreme points 9

6 Optimality of extreme points 9

References 10

* Based heavily on Chapter 2 in [1]

1
1 Polyhedra
We define a polyhedron (plural: polyhedra) as follows:

Definition 1.1 (polyhedron). A polyhedron is a set that can be described in the form

{x ∈ Rn : Ax ≥ b}

where A is an m × n matrix and b is a vector in Rm .

We also note that a set of the form

{x ∈ Rn : Ax = b, x ≥ 0}

is a polyhedron but written in standard form.

Definition 1.2 (bounded set). A set S ⊆ Rn is bounded if there exists a constant K s.t. the
absolute value of every component of every element of S is less than or equal to K.

Definition 1.3 (hyperplane/halfspace). Let a be a nonzero vector in Rn and let b be a scalar.

1. The set {x ∈ Rn : aT x = b} is called a hyperplane.

2. The set {x ∈ Rn : aT x ≥ b} is called a halfspace.

Note that a hyperplane is the boundary of a corresponding halfspace. Also, the vector a in the
definition of the hyperplane is perpendicular to the hyperplane itself. One can see this by taking
points x, y belonging to the same hyperplane and observing that aT x = aT y =⇒ aT (x − y) = 0.
This implies that a is orthogonal to any direction vector confined to the hyperplane.
Observe that Ax ≥ b is equivalent to requiring aTi x ≥ bi for each i ∈ [m], where ai are the
rows of A. It follows that a polyhedron is the intersection of finitely many halfspaces.

2 Extreme points, vertices, and basic feasible solutions

Definition 2.1 (extreme point). Let P be a polyhedron. A vector x ∈ P is an extreme point of P
if we cannot find two vectors y, z ∈ P both different from x, and a scalar λ ∈ [0, 1] such that

x = λy + (1 − λ)z

That is to say, if x lies on any line segment within P , it can only be an endpoint of that line
segment. Or more succinctly, x ∈ / relint[y, z] for any y, z ∈ P distinct from x.

Definition 2.2 (vertex). Let P be a polyhedron. A vector x ∈ P is a vertex of P if there exists

some c ∈ Rn s.t. cT x < cT y for all y satisfying y ∈ P, y ̸= x.

Geometrically, this is saying that x is a vertex of P iff we can construct a hyperplane with P on
one side of it s.t. it meets P only at the point x. Another way to think of this is that we can balance
a hyperplane on the point x in a way that the hyperplane does not touch any other point in P .

2
 The above two definitions are too generic however to be useful in the design of algorithms. We
want a definition that relies on a representation of a polyhedron in terms of linear constraints, and
which reduces to an algebraic test. Consider a polyhedron P ⊆ Rn defined in terms of the linear
equality and inequality constraints:

aTi x ≥ bi , i ∈ M1 ,
aTi x ≤ bi , i ∈ M2 ,
aTi x = bi , i ∈ M3 ,

where M1 , M2 and M3 are finite index sets, each ai is a vector in Rn , and each bi is a scalar.
Definition 2.3 (active constraint). If a vector x∗ satisfies aTi x∗ = bi for some i ∈ M1 , M2 or M3 ,
we say that the corresponding constraint is active at x∗ .
Observe that if there are n constraints that are active at a vector x∗ , then x∗ satisfies a certain
system of n linear equations in n unknowns. This system has a unique solution iff these n equations
are linearly independent.
Theorem 2.4. Let x∗ be an element of Rn and let I = {i : aTi x = bi } be the set of indices of
constraints that are active at x∗ . Then, TFAE:
1. There exist n vectors in the set {ai : i ∈ I} which are linearly independent.

2. The span of the vectors ai where i ∈ I, is all of of Rn .

3. The systems of equations aTi x = bi where i ∈ I, has a unique solution.

Proof. Suppose that the vectors ai , i ∈ I span Rn . Then the span of these vectors has dimension n.
Hence, n of these vectors form a basis of Rn , and are therefore linearly independent. Conversely,
suppose that n of the vectors ai , i ∈ I are linearly independent. Then, the subspace spanned by
these n vectoes is n-dimensional, so must be equal to Rn . We have shown the equivalence of 1.
and 2.
Now suppose the system of equations aTi x = bi , i ∈ I has multiple solutions, say x1 and x2 .
Then the nonzero vector d = x1 − x2 satisfies aTi d = 0 for all i ∈ I. Since d is orthogonal
to every vector ai , i ∈ I, d cannot be a linear combination of these vectors, so ai , i ∈ I cannot
span Rn . Conversely, if ai , i ∈ I do not span Rn , choose a nonzero vector d which is orthogonal
to the subspace spanned by these vectors. If x satisfies aTi x = bi for all i, we also must have
aTi (x + d) = bi for all i. This proves the equivalence of 2. and 3.
Note. If we say that certain constraints are linearly independent, we means that the corresponding
vectors ai are linearly independent.
We now give an algebraic definition of a ”corner point” as a feasible solution at which there
are n linearly independent active constraints.
Definition 2.5 (basic solution). Consider a polyhedron P defined by linear equality and inequality
constraints, and let x∗ be an element of Rn .
1. The vector x∗ is a basic solution if:

3
 (a) All equality constraints are active;
(b) Out of the constraints that are active at x∗ , there are n of them that are linearly inde-
pendent.
2. If x∗ is a basic solution that satisfies all of the constraints, we say it is a basic feasible
solution (BFS).
Suppose we want to look for basic solutions of a polyhedron. Of course, we can first impose
the equality constraints for feasibility. Then we can require that enough additional constraints are
active, so we get a total of n linearly independent active constraints. Once we have n linearly
independent active constraints, by Theorem 2.4 a unique vector x∗ is determined. However, this
procedure provides no guarantee on the feasibility of x∗ since some of the inactive constraints
could be violated. In the case that some of the inactive constraints are violated, we have a basic
(but not basic feasible) solution.
Theorem 2.6. Let P be a nonempty polyhedron and let x∗ ∈ P . Then TFAE:
1. x∗ is a vertex;
2. x∗ is an extreme point;
3. x∗ is a basic feasible solution.
Proof. For the purposes of this proof, we assume wlog that P is represented in terms of constraints
of the form aTi x ≥ bi and aTi x = bi .
Suppose x∗ ∈ P is a vertex. By definition, there exists some c ∈ Rn such that cT x∗ < cT y
for every y ∈ P with y ̸= x∗ . If y ∈ P, z ∈ P , y ̸= x∗ , z ̸= x∗ , and λ ∈ [0, 1], cT x∗ < cT y and
cT x∗ < cT z. It follows that cT x∗ < cT (λy + (1 − λ)z) and, therefore, x∗ ̸= λy + (1 − λ)z. Thus,
x∗ is an extreme point.
Suppose x∗ ∈ P is not a basic feasible solution and let I = {i : aTi x∗ = bi }. Since x∗ is not
a BFS, there do not exist n linearly independent vectors in the family ai , i ∈ I. Thus, the vectors
ai , i ∈ I, lie in a proper subspace of Rn , and there exists some nonzero vector d ∈ Rn such that
aTi d = 0, for all i ∈ I. Let ϵ > 0 be sufficiently small and consider the vectors y = x∗ + ϵd and
z = x∗ − ϵd. Notice that aTi y = aTi x∗ = bi , for i ∈ I. Furthermore, for i ∈ / I, we have aTi x∗ > bi
and, provided that ϵ is small enough, we will also have aTi y > bi (it suffices to choose ϵ so that
ϵ|aTi d| < aTi x∗ − bi for all i ∈
/ I). Thus, when ϵ is small enough, y ∈ P and, by a similar argument,
z ∈ P . We finally notice that x∗ = (y + z)/2, which implies that x∗ is not an extreme point.
Let x∗ be a BFS and let I = {i : aTi x∗ = bi }. Let c = i∈I ai . We then have
P

X X
cT x ∗ = aTi x∗ = bi
i∈I i∈I

Furthermore, for any x ∈ P and any i, we have aTi x ≥ bi , and

X X
cT x = aTi x ≥ bi (2.1)
i∈I i∈I

This shows that x∗ is an optimal solution to the problem of minimizing cT x over the set P . Fur-
thermore, equality holds in (2.1) iff aTi x = bi for all i ∈ I. Since x∗ is a BFS, there are n linearly

4
independent constraints that are active at x∗ , and x∗ is the unique solution t the system of equations
aTi x = bi , i ∈ I. It follows that x∗ is the unique minimizer cT x over the set P and, therefore, x∗ is
a vertex of P .
Corollary 2.7. Given a finite number of linear inequality constraints, there can only be a finite
nunber of basic or basic feasible solutions.
Proof. Consider a system of m linear inequality constraints imposed on a vector x ∈ Rn . At any
basic solution, there are n linearly independent active constraints. Since any n linearly independent
active constraints define a unique point, it follows that different basic solutions correspond to
different sets of n linearly independent active constraints. Therefore, the number of basic solutions
is bounded above by the number of ways that we can choose n constraints out of a total m, which
is finite.
Although the number of basic, and therefore, basic feasible solutions is guaranteed to be finite,
it can be very large. For example, take the unit cube:

{x ∈ Rn : 0 ≤ xi ≤ 1, i ∈ [n]}

The unit cube is defined in terms of 2n constraints (n for 0 ≤ xi and n for xi ≤ 1), but it has 2n
basic feasible solutions (the corners of the unit cube).
Definition 2.8 (adjacent basic solution). Two distinct basic solutions to a set of linear constraints
in Rn are said to be adjacent if we can find n − 1 linearly independent constraints that are active
at both of them. If two adjacent basic solutions are also feasible, then the line segment that joins
them is called an edge of the feasible set.

3 Polyhedra in standard form

Consider the aforementioned standard form polyhedron

P = {x ∈ Rn : Ax = b, x ≥ 0}

Let A have dimensions m × n and note m is the number of equality constraints. We will make
the assumption that the m rows of the matrix are linearly independent. Since the rows are n-
dimensional, this requires that m ≤ n.
We will see that when P is nonempty, linearly dependent rows of A correspond to redundant
constraints that can be discarded; therefore our linear independence assumption can be made with-
out loss of generality.
Recall that at any basic solution, there must be n linearly independent constraints that are active.
We know that every basic solution satisfies the equality constraints Ax = b, which provides us with
m active constraints. Because of our assumption on the rows of A, these are linearly independent.
In order to obtain a total of n active constraints, we need to choose n − m of the variables xi and
set them to zero. This makes the constraints xi ≥ 0 active.
Theorem 3.1. Consider the constraints Ax = b and x ≥ 0. Assume that the m × n matrix A has
linearly independent rows. A vector x ∈ Rn is a basic solution iff we have Ax = b, and there exist
indices B(1), . . . , B(m) such that

5
 1. The columns of AB(1) , . . . , AB(m) are linearly indpendent.

2. If i ̸= B(1), . . . , B(m), then xi = 0.

Proof. See [1, Theorem 2.4].

In view of Theorem 3.1, all basic solutions ot a standard form polyhedron can be constructed
according to the following procedure:

1. Choose m linearly independent columns AB(1) , . . . , AB(m) .

2. Let xi = 0 for all i ̸= B(1), . . . , B(m).

3. Solve the system of m equations Ax = b for the unknowns xB(1) , . . . , xB(m) .

If a basic solution constructed according to this procedure is nonnegative, then it is feasible, and it
is a basic feasible solution. Conversely, since every basic feasible solution is a basic solution, any
BFS can be obtained from this procedure.
Notation. If x is a basic solution, the variables xB(1) , . . . , xB(m) are called basic variables; the
remaining variables are called nonbasic. The columns AB(1) , . . . , AB(m) are called the basic
columns and, since they are linearly independent, they form a basis of Rm .
We say two bases are distinct if they have different sets {B(1), . . . , B(m)} of basic indices;
if two bases involve the same set of indices in a different order, they will be considered the same
basis.
If we arrange the m basic variables next to each other in a matrix, we obtain an m × n matrix
B, called the basis matrix. This matrix must be invertible, since we require the basic columns to
be linearly independent. We can similarly define a vector xB with the values of the basic variables.
Thus,  
xB(1)
B = AB(1) AB(2) . . . AB(m) , xB =  ...  .
   
xB(m)
We can determine the basic variables by solving the equation BxB = b. Observe this system of
equations has a unique solution, since B has linearly independent columns.

Example 3.2. Let the constraint Ax = b be of the form

   
1 1 2 1 0 0 0 8
0 1 6 0 1 0 0 12
1 0 0 0 0 1 0 x =  4 
   

0 1 0 0 0 0 1 6

Let us choose A4 , A5 , A6 , A7 as our basic columns. Note that they are linearly independent and the
corresponding basis matrix is the identity. We then have the basic solution x = (0, 0, 0, 8, 12, 4, 6)
which is nonnegative and therefore a BFS. Another basis can be obtained by choosing the columns
A3 , A5 , A6 , A7 (note that they are linearly independent). The corresponding basic solution is x =
(0, 0, 4, 0, −12, 4, 6), which is not feasible because x5 < 0.

6
 Suppose now that there was an eighth column A8 , identical to A7 . Then the two sets of columns
{A3 , A5 , A6 , A7 } and {A3 , A5 , A6 , A8 } coincide. On the other hand, the corresponding sets of
basic indices, which are {3, 5, 6, 7} and {3, 5, 6, 8}, are different and we have two different bases,
according to our conventions.
Note. Different basic solutions must correspond to different bases, because a basis uniquely de-
termines a basic solution. However, two different bases may lead to the same basic solution. In
other words, if we have two different basic solutions, they cannot have come from the same basis.
However, it is possible that two different bases give rise to the same basic solution.
Definition 3.3 (adjacent basis). We say that two bases are adjacent if they share all but one basic
column.
It is not hard to check that adjacent basic solutions can always be obtained from two adjacent
bases. Conversely, if two adjacent bases lead to distinct basic solutions, then the two basic solutions
are adjacent.
Example 3.4. Referring back to Example 3.2, the bases {A4 , A5 , A6 , A7 } and {A3 , A5 , A6 , A7 } are
adjacent. The corresponding basic solutions x = (0, 0, 0, 8, 12, 4, 6) and x = (0, 0, 4, 0, −12, 4, 6).
are adjacent since n = 7 and we have a total of six common linearly independent active constraints;
these are x1 ≥ 0, x2 ≥ 0, and the four equality constraints.
The following theorem will show that with no loss of generality, we may assume that A has
foll row rank.
Theorem 3.5. Let P = {x : Ax = b, x ≥ 0} be a nonempty polyhedron, where A is a matrix
of dimensions m × n, with rows aT1 , . . . , aTm . Suppose that rank(A) = k < m and that the rows
aTi1 , . . . , aTik are linearly independent. Consider the polyhedron

Q = {x : aTi1 x = bi1 , . . . , aTik = bik , x ≥ 0}

Then Q = P .
Proof. See [1, Theorem 2.5].
Notice that the polyhedron Q in the theorem above is in standard form: namely, Q = {x :
Dx = f, x ≥ 0}, where D is a k × n submatrix of A, with rank equal to k, and f is a k-
dimensional subvector of b. We conclude that as long as the feasible set is nonempty, a linear
programming problem in standard form can be reduced to an equivalent standard form problem.
Example 3.6. Consider the (nonempty) polyhedron defined by the constraints
2x1 + x2 + x3 = 2
x 1 + x2 =1
x1 + x3 = 1
x1 , x2 , x3 ≥ 0
Observe that rank(A) = 2, since the last two rows (1, 1, 0) and (1, 0, 1) are linearly independent,
but the first row is equal to the sum of the other two rows. So the first constraint is redundant and
after it is removed, we still have the same polyhedron.

7
4 Degeneracy
With the definition of a basic solution that we are using, we must have n linearly independent
active constraints. This allows for the possibility that the number of active constraints is greater
than n. Of course, in n dimensions, no more than n of them can be linearly independent. In this
case, we say we have a degenerate basic solution. In other words, at a degenerate basic solution,
the number of active constraints is greater than the minimum necessary.

Definition 4.1 (degenerate basic solution). A basic solution x ∈ Rn is said to be degenerate if

more than n of the constraints are active at x.

Example 4.2. Consider the polyhedron P defined by the constraints

x1 + x2 + 2x3 ≤8
x2 + 6x3 ≤ 12
x1 ≤4
x2 ≤6
x1 , x2 , x3 ≥0

The vector x = (2, 6, 0) is a nondegenerate basic feasible solution, because there are exactly three
active and linearly independent constraints, namely x1 + x2 + 2x3 ≤ 8, x2 ≤ 6, and x3 ≥ 0. The
vector x = (4, 0, 2) is a degenerate basic feasible solution, because there are four active constraints,
three of them linearly independent, namely x1 + x2 + 2x3 ≤ 8, x2 + 6x3 ≤ 12, x1 ≤ 4, and x2 ≥ 0.

For standard form polyhedra, we have the following definition of degeneracy:

Definition 4.3. Consider the standard form polyhedra P = {x ∈ Rn : Ax = b, x ≥ 0} and let x

be a basic solution. Let m be the number of rows A. The vector x is a degenerate basic solution if
more than n − m of the components of x are zero.

Essentially what is happening in degeneracy is that we pick a basic solution by picking n

linearly independent constraints to be satisfied with equality, and we realize that certain other
constraints are also satisfied with equality.

Observation 1. Degeneracy is not necessarily representation independent; it may depend on the

particular representation of a polyhedron. For example, consider a nondegenerate BFS x∗ of a
standard form polyhedron P = {x : Ax = b, x ≥ 0}, where A is m×n. We know by Theorem 3.1,
n − m of the variables x∗i are equal to zero. Let us now represent P in the form

P = {x : Ax ≥ b, −Ax ≥ −b, x ≥ 0}.

Now at the BFS x∗ , we have n−m variables set to zero and an additional 2m inequality constraints
satisfied with equality. We therefore have n + m active constraints and x∗ is degenerate. It follows
that every BFS is degenerate under the second representation.
Still, it can be shown that if a BFS is degenerate under one particular standard form represen-
tation, then it is degenerate under every standard form representation of the same polyhedron.

8
5 Existence of extreme points
Definition 5.1 (containing a line). A polyhedron P ⊆ Rn contains a line if there exists a vector
x ∈ P and a nonzero vector d ∈ Rn such that x + λd ∈ P for all scalars λ.
Theorem 5.2. Suppose the polyhedron P = {x ∈ Rn : aTi x ≥ b, i = 1, . . . , m} is nonempty.
Then, TFAE:
1. The polyhedron P has at least one extreme point.

2. The polyhedron P does not contain a line.

3. There exist n vectors out of the family a1 , . . . , an , which are linearly independent.
Proof. See [1, Theorem 2.6]
Note. A bounded polyhedron does not contain a line. Similarly, the positive orthant does not
contain a line. Since a polyhedron in standard form is contained in the positive orthant, it does not
contain a line either.
Corollary 5.3. Every nonempty bounded polyhedron and every nonempty polyhedron in standard
form has at least one basic feasible solution.

6 Optimality of extreme points

Theorem 6.1. Consider the linear programming problem of minimizing cT x over a polyhedron P .
Suppose that P has at least one extreme point and that there exists an optimal solution. Then, there
exists an optimal solution which is an extreme point of P .
Proof. Let Q be the set of all optimal solutions, which we have assumed to be nonempty. Let P
be of the form P = {x ∈ Rn : Ax ≥ b} and let v be the optimal value of the cost cT x. Then
Q = {x ∈ Rn : Ax ≥ b, cT x = v}, which is clearly a polyhedron as well. Since Q ⊆ P , and since
P contains no lines (Theorem 5.2), Q contains no lines either. Therefore, Q has an extreme point.
Let x∗ be an extreme point of Q. We will show that x∗ is also an extreme point of P . Suppose for
contradiction that x∗ is not an extreme point of P . Then, there exists y, z ∈ P distinct from x∗ ,
and some λ ∈ [0, 1] such that x∗ = λy + (1 − λ)z. It follows that v = cT x∗ = λcT y + (1 − λ)cT z.
Furthermore, since v is the optimal cost, cT y ≥ v and cT z ≥ v. This implies that cT y = cT z = v
and therefore, z, y ∈ Q. But this is a contradiction because x∗ is an extreme point of Q. It follows
that x∗ is an extreme point of P . In addition, since x∗ belongs to Q, it is optimal.
The above theorem applies to polyhedra in standard form, as well as to bounded polyhedra,
since they do not contain a line.
The next result is stronger than Theorem 6.1. It shows that the existence of an optimal solution
can be taken for granted, as long as the optimal cost is finite.
Theorem 6.2. Consider the linear programming problem of minimizing cT x over a polyhedron P .
Suppose that P has at least one extreme point. Then, either the optimal cost is equal to −∞, or
there exists an extreme point which is optimal.

9
Proof. The proof idea is very similar to that of Theorem 6.1. See [1, Theorem 2.8].
For a general linear programming problem, if the feasible set has no extreme points, then
Theorem 6.2 does not apply directly. On the other hand, any linear programming problem can be
transformed into an equivalent problem in standard form. This establishes the following corollary.

Corollary 6.3. Consider the linear programming problem of minimizing cT x over a nonempty
polyhedron. Then, either the optimal cost is equal to −∞ or there exists an optimal solution.

References
[1] D. B ERTSIMAS AND J. T SITSIKLIS, Introduction to Linear Optimization, Athena Scientific,
1997.

10