CS 296.

1
Mathematical Modelling of Continuous Systems
Carlo Tomasi
Duke University
Fall 2004
2
Chapter 1
Introduction
Fields such as robotics or computer vision are interdisciplinary subjects at the intersection of engineering and computer
science. By their nature, they deal with both computers and the physical world. Although the former are in the latter,
the workings of computers are best described in the black-and-white vocabulary of discrete mathematics, which is
foreign to most classical models of reality, quantum physics notwithstanding.
This class surveys some of the key tools of applied math to be used at the interface of continuous and discrete. It
is not on robotics or computer vision, nor does it cover any other application area. Applications evolve rapidly, but
their mathematical foundations remain. Even if you will not pursue any of these fields, the mathematics that you learn
in this class will not go wasted. To be sure, applied mathematics is a discipline in itself and, in many universities, a
separate department. Consequently, this class can be a quick tour at best. It does not replace calculus or linear algebra,
which are assumed as prerequisites, nor is it a comprehensive survey of applied mathematics. What is covered is a
compromise between the time available and what is useful and fun to talk about. Even if in some cases you may have
to wait until you take an applied class to fully appreciate the usefulness of a particular topic, I hope that you will enjoy
studying these subjects in their own right.
1.1 Who Should Take This Class
The main goal of this class is to present a collection of mathematical tools for both understanding and solving problems
in fields that manipulate models of the real world, such as robotics, artificial intelligence, vision, engineering, or several
aspects of the biological sciences. Several classes at most universities each cover some of the topics presented in this
class, and do so in much greater detail. If you want to understand the full details of any one of the topics in the
syllabus below, you should take one or more of these other classes instead. If you want to understand how these tools
are implemented numerically, you should take one of the classes in the scientific computing program, which again
cover these issues in much better detail. Finally, if you want to understand robotics, vision, or other applied fields, you
should take classes in these subjects, since this course is not on applications.
On the other hand, if you do plan to study robotics, vision, or other applied subjects in the future, and you regard
yourself as a user of the mathematical techniques outlined in the syllabus below, then you may benefit from this course.
Of the proofs, we will only see those that add understanding. Of the implementation aspects of algorithms that are
available in, say, Matlab or LApack, we will only see the parts that we need to understand when we use the code.
In brief, we will be able to cover more topics than other classes because we will be often (but not always) un-
concerned with rigorous proof or implementation issues. The emphasis will be on intuition and on practicality of the
various algorithms. For instance, why are singular values important, and how do they relate to eigenvalues? What are
the dangers of Newton-style minimization? How does a Kalman filter work, and why do PDEs lead to sparse linear
systems? In this spirit, for instance, we discuss Singular Value Decomposition and Schur decomposition both because
they never fail and because they clarify the structure of an algebraic or a differential linear problem.
3
4 CHAPTER 1. INTRODUCTION
1.2 Syllabus
Here is the ideal syllabus, but how much we cover depends on how fast we go.
1. Introduction
2. Unknown numbers
2.1 Algebraic linear systems
2.1.1 Characterization of the solutions to a linear system
2.1.2 Gaussian elimination
2.1.3 The Singular Value Decomposition
2.1.4 The pseudoinverse
2.2 Function optimization
2.2.1 Newton and Gauss-Newton methods
2.2.2 Levenberg-Marquardt method
2.2.3 Constraints and Lagrange multipliers
3. Unknown functions of one real variable
3.1 Ordinary differential linear systems
3.1.1 Eigenvalues and eigenvectors
3.1.2 The Schur decomposition
3.1.3 Ordinary differential linear systems
3.1.4 The matrix zoo
3.1.5 Real, symmetric, positive-definite matrices
3.2 Statistical estimation
3.2.1 Linear estimation
3.2.2 Weighted least squares
3.2.3 The Kalman filter
4. Unknown functions of several variables
4.1 Tensor fields of several variables
4.1.1 Grad, div, curl
4.1.2 Line, surface, and volume integrals
4.1.3 Green’s theorem and potential fields of two variables
4.1.4 Stokes’ and divergence theorems and potential fields of three variables
4.1.5 Diffusion and flow problems
4.2 Partial differential equations and sparse linear systems
4.2.1 Finite differences
4.2.2 Direct versus iterative solution methods
4.2.3 Jacobi and Gauss-Seidel iterations
4.2.4 Successive overrelaxation
4.3 Calculus of variations
4.3.1 Euler-Lagrange equations
4.3.2 The brachistochrone
1.3. DISCUSSION OF THE SYLLABUS 5
1.3 Discussion of the Syllabus
In robotics, vision, physics, and any other branch of science whose subject belongs to or interacts with the real
world, mathematical models are developed that describe the relationship between different quantities. Some of these
quantities are measured, or sensed, while others are inferred by calculation. For instance, in computer vision, equations
tie the coordinates of points in space to the coordinates of corresponding points in different images. Image points are
data, world points are unknowns to be computed.
Similarly, in robotics, a robot arm is modeled by equations that describe where each link of the robot is as a
function of the configuration of the link’s own joints and that of the links that support it. The desired position of the
end effector, as well as the current configuration of all the joints, are the data. The unknowns are the motions to be
imparted to the joints so that the end effector reaches the desired target position.
Of course, what is data and what is unknown depends on the problem. For instance, the vision system mentioned
above could be looking at the robot arm. Then, the robot’s end effector position could be the unknowns to be solved
for by the vision system. Once vision has solved its problem, it could feed the robot’s end-effector position as data for
the robot controller to use in its own motion planning problem.
Sensed data are invariably noisy, because sensors have inherent limitations of accuracy, precision, resolution,
and repeatability. Consequently, the systems of equations to be solved are typically overconstrained: there are more
equations than unknowns, and it is hoped that the errors that affect the coefficients of one equation are partially
cancelled by opposite errors in other equations. This is the basis of optimization problems: Rather than solving a
minimal system exactly, an optimization problem tries to solve many equations simultaneously, each of them only
approximately, but collectively as well as possible, according to some global criterion. Least squares is perhaps the
most popular such criterion, and we will devote a good deal of attention to it.
In summary, the problems encountered in robotics and vision, as well as other applications of mathematics, are
optimization problems. A fundamental distinction between different classes of problems reflects the complexity of the
unknowns. In the simplest case, unknowns are scalars. When there is more than one scalar, the unknown is a vector
of numbers, typically either real or complex. Accordingly, the first part of this course will be devoted to describing
systems of algebraic equations, especially linear equations, and optimization techniques for problems whose solution
is a vector of reals. The main tool for understanding linear algebraic systems is the Singular Value Decomposition
(SVD), which is both conceptually fundamental and practically of extreme usefulness. When the systems are nonlinear,
they can be solved by various techniques of function optimization, of which we will consider the basic aspects.
Since physical quantities often evolve over time, many problems arise in which the unknowns are themselves
functions of time. This is our second class of problems. Again, problems can be cast as a set of equations to be solved
exactly, and this leads to the theory of Ordinary Differential Equations (ODEs). Here, “ordinary” expresses the fact
that the unknown functions depend on just one variable (e.g., time). The main conceptual tool for addressing ODEs is
the theory of eigenvalues, and the primary computational tool is the Schur decomposition.
Alternatively, problems with time varying solutions can be stated as minimization problems. When viewed glob-
ally, these minimization problems lead to the calculus of variations. When the minimization problems above are
studied locally, they become state estimation problems, and the relevant theory is that of dynamic systems and Kalman
filtering.
The third category of problems concerns unknown functions of more than one variable. The images taken by a
moving camera, for instance, are functions of time and space, and so are the unknown quantities that one can compute
fromthe images, such as the distance of points in the world fromthe camera. This leads to Partial Differential equations
(PDEs), or to extensions of the calculus of variations. In this class, we will see how PDEs arise, and how they can be
solved numerically.
6 CHAPTER 1. INTRODUCTION
1.4 Books
The class will be based on these lecture notes, and additional notes handed out when necessary. Other useful references
include the following.
• R. Courant and D. Hilbert, Methods of Mathematical Physics, Volume I and II, John Wiley and Sons, 1989.
• D. A. Danielson, Vectors and Tensors in Engineering and Physics, Addison-Wesley, 1992.
• J. W. Demmel, Applied Numerical Linear Algebra, SIAM, 1997.
• A. Gelb et al., Applied Optimal Estimation, MIT Press, 1974.
• P. E. Gill, W. Murray, and M. H. Wright, Practical Optimization, Academic Press, 1993.
• G. H. Golub and C. F. Van Loan, Matrix Computations, 2nd Edition, Johns Hopkins University Press, 1989, or
3rd edition, 1997.
• W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes in C, 2nd Edition,
Cambridge University Press, 1992.
• G. Strang, Introduction to Applied Mathematics, Wellesley- Cambridge Press, 1986.
• A. E. Taylor and W. R. Mann, Advanced Calculus, 3rd Edition, John Wiley and Sons, 1983.
• L. N. Trefethen and D. Bau, III, Numerical Linear Algebra, SIAM, 1997.
• R. Weinstock, Calculus of Variations, Dover, 1974.
Chapter 2
Algebraic Linear Systems
An algebraic linear system is a set of m equations in n unknown scalars, which appear linearly. Without loss of
generality, an algebraic linear system can be written as follows:
Ax = b (2.1)
where A is an m× n matrix, x is an n-dimensional vector that collects all of the unknowns, and b is a known vector
of dimension m. In this chapter, we only consider the cases in which the entries of A, b, and x are real numbers.
Two reasons are usually offered for the importance of linear systems. The first is apparently deep, and refers
to the principle of superposition of effects. For instance, in dynamics, superposition of forces states that if force
f
1
produces acceleration a
1
(both possibly vectors) and force f
2
produces acceleration a
2
, then the combined force
f
1
+ αf
2
produces acceleration a
1
+ αa
2
. This is Newton’s second law of dynamics, although in a formulation less
common than the equivalent f = ma. Because Newton’s laws are at the basis of the entire edifice of Mechanics,
linearity appears to be a fundamental principle of Nature. However, like all physical laws, Newton’s second law is an
abstraction, and ignores viscosity, friction, turbulence, and other nonlinear effects. Linearity, then, is perhaps more in
the physicist’s mind than in reality: if nonlinear effects can be ignored, physical phenomena are linear!
A more pragmatic explanation is that linear systems are the only ones we know how to solve in general. This
argument, which is apparently more shallow than the previous one, is actually rather important. Here is why. Given
two algebraic equations in two variables,
f(x, y) = 0
g(x, y) = 0 ,
we can eliminate, say, y and obtain the equivalent system
F(x) = 0
y = h(x) .
Thus, the original system is as hard to solve as it is to find the roots of the polynomial F in a single variable. Unfortu-
nately, if f and g have degrees d
f
and d
g
, the polynomial F has generically degree d
f
d
g
.
Thus, the degree of a system of equations is, roughly speaking, the product of the degrees. For instance, a system of
m quadratic equations corresponds to a polynomial of degree 2
m
. The only case in which the exponential is harmless
is when its base is 1, that is, when the system is linear.
In this chapter, we first review a few basic facts about vectors in sections 2.1 through 2.4. More specifically, we
develop enough language to talk about linear systems and their solutions in geometric terms. In contrast with the
promise made in the introduction, these sections contain quite a few proofs. This is because a large part of the course
material is based on these notions, so we want to make sure that the foundations are sound. In addition, some of the
proofs lead to useful algorithms, and some others prove rather surprising facts. Then, in section 2.5, we characterize
the solutions of linear algebraic systems.
7
8 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
2.1 Linear (In)dependence
Given n vectors a
1
, . . . , a
n
and n real numbers x
1
, . . . , x
n
, the vector
b =
n

j=1
x
j
a
j
(2.2)
is said to be a linear combination of a
1
, . . . , a
n
with coefficients x
1
, . . . , x
n
.
The vectors a
1
, . . . , a
n
are linearly dependent if they admit the null vector as a nonzero linear combination. In
other words, they are linearly dependent if there is a set of coefficients x
1
, . . . , x
n
, not all of which are zero, such that
n

j=1
x
j
a
j
= 0 . (2.3)
For later reference, it is useful to rewrite the last two equalities in a different form. Equation (2.2) is the same as
Ax = b (2.4)
and equation (2.3) is the same as
Ax = 0 (2.5)
where
A =
_
a
1
· · · a
n
¸
, x =
_
¸
_
x
1
.
.
.
x
n
_
¸
_ , b =
_
¸
_
b
1
.
.
.
b
m
_
¸
_ .
If you are not convinced of these equivalences, take the time to write out the components of each expression for a
small example. This is important. Make sure that you are comfortable with this.
Thus, the columns of a matrix A are dependent if there is a nonzero solution to the homogeneous system (2.5).
Vectors that are not dependent are independent.
Theorem 2.1.1 The vectors a
1
, . . . , a
n
are linearly dependent iff
1
at least one of them is a linear combination of the
others.
Proof. In one direction, dependency means that there is a nonzero vector x such that
n

j=1
x
j
a
j
= 0 .
Let x
k
be nonzero for some k. We have
n

j=1
x
j
a
j
= x
k
a
k
+
n

j=1, j=k
x
j
a
j
= 0
so that
a
k
= −
n

j=1, j=k
x
j
x
k
a
j
(2.6)
as desired. The converse is proven similarly: if
a
k
=
n

j=1, j=k
x
j
a
j
1
“iff” means “if and only if.”
2.2. BASIS 9
for some k, then
n

j=1
x
j
a
j
= 0
by letting x
k
= −1 (so that x is nonzero). ∆
2
We can make the first part of the proof above even more specific, and state the following
Lemma 2.1.2 If n nonzero vectors a
1
, . . . , a
n
are linearly dependent then at least one of them is a linear combination
of the ones that precede it.
Proof. Just let k be the last of the nonzero x
j
. Then x
j
= 0 for j > k in (2.6), which then becomes
a
k
=
n

j<k
x
j
x
k
a
j
as desired. ∆
2.2 Basis
A set a
1
, . . . , a
n
is said to be a basis for a set B of vectors if the a
j
are linearly independent and every vector in B can
be written as a linear combination of them. B is said to be a vector space if it contains all the linear combinations of
its basis vectors. In particular, this implies that every linear space contains the zero vector. The basis vectors are said
to span the vector space.
Theorem 2.2.1 Given a vector b in the vector space B and a basis a
1
, . . . , a
n
for B, the coefficients x
1
, . . . , x
n
such
that
b =
n

j=1
x
j
a
j
are uniquely determined.
Proof. Let also
b =
n

j=1
x

j
a
j
.
Then,
0 = b −b =
n

j=1
x
j
a
j

n

j=1
x

j
a
j
=
n

j=1
(x
j
−x

j
)a
j
but because the a
j
are linearly independent, this is possible only when x
j
−x

j
= 0 for every j. ∆
The previous theorem is a very important result. An equivalent formulation is the following:
If the columns a
1
, . . . , a
n
of A are linearly independent and the system Ax = b admits a solution, then
the solution is unique.
2
This symbol marks the end of a proof.
10 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
Pause for a minute to verify that this formulation is equivalent.
Theorem 2.2.2 Two different bases for the same vector space B have the same number of vectors.
Proof. Let a
1
, . . . , a
n
and a

1
, . . . , a

n
be two different bases for B. Then each a

j
is in B (why?), and can therefore
be written as a linear combination of a
1
, . . . , a
n
. Consequently, the vectors of the set
G = a

1
, a
1
, . . . , a
n
must be linearly dependent. We call a set of vectors that contains a basis for B a generating set for B. Thus, G is a
generating set for B.
The rest of the proof now proceeds as follows: we keep removing a vectors from G and replacing them with a

vectors in such a way as to keep Ga generating set for B. Then we show that we cannot run out of a vectors before we
run out of a

vectors, which proves that n ≥ n

. We then switch the roles of a and a

vectors to conclude that n

≥ n.
This proves that n = n

.
From lemma 2.1.2, one of the vectors in G is a linear combination of those preceding it. This vector cannot be a

1
,
since it has no other vectors preceding it. So it must be one of the a
j
vectors. Removing the latter keeps Ga generating
set, since the removed vector depends on the others. Now we can add a

2
to G, writing it right after a

1
:
G = a

1
, a

2
, . . . .
G is still a generating set for B.
Let us continue this procedure until we run out of either a vectors to remove or a

vectors to add. The a vectors
cannot run out first. Suppose in fact per absurdum that G is now made only of a

vectors, and that there are still
left-over a

vectors that have not been put into G. Since the a

s form a basis, they are mutually linearly independent.
Since B is a vector space, all the a

s are in B. But then G cannot be a generating set, since the vectors in it cannot
generate the left-over a

s, which are independent of those in G. This is absurd, because at every step we have made
sure that G remains a generating set. Consequently, we must run out of a

s first (or simultaneously with the last a).
That is, n ≥ n

.
Now we can repeat the whole procedure with the roles of a vectors and a

vectors exchanged. This shows that
n

≥ n, and the two results together imply that n = n

. ∆
A consequence of this theorem is that any basis for R
m
has m vectors. In fact, the basis of elementary vectors
e
j
= jth column of the m×m identity matrix
is clearly a basis for R
m
, since any vector
b =
_
¸
_
b
1
.
.
.
b
m
_
¸
_
can be written as
b =
m

j=1
b
j
e
j
and the e
j
are clearly independent. Since this elementary basis has m vectors, theorem 2.2.2 implies that any other
basis for R
m
has m vectors.
Another consequence of theorem 2.2.2 is that n vectors of dimension m < n are bound to be dependent, since any
basis for R
m
can only have m vectors.
Since all bases for a space have the same number of vectors, it makes sense to define the dimension of a space as
the number of vectors in any of its bases.
2.3. INNER PRODUCT AND ORTHOGONALITY 11
2.3 Inner Product and Orthogonality
In this section we establish the geometric meaning of the algebraic notions of norm, inner product, projection, and
orthogonality. The fundamental geometric fact that is assumed to be known is the law of cosines: given a triangle with
sides a, b, c (see figure 2.1), we have
a
2
= b
2
+c
2
−2bc cos θ
where θ is the angle between the sides of length b and c. A special case of this law is Pythagoras’ theorem, obtained
when θ = ±π/2.
θ
c
b
a
Figure 2.1: The law of cosines states that a
2
= b
2
+c
2
−2bc cos θ.
In the previous section we saw that any vector in R
m
can be written as the linear combination
b =
m

j=1
b
j
e
j
(2.7)
of the elementary vectors that point along the coordinate axes. The length of these elementary vectors is clearly one,
because each of them goes from the origin to the unit point of one of the axes. Also, any two of these vectors form a
90-degree angle, because the coordinate axes are orthogonal by construction. How long is b? From equation (2.7) we
obtain
b = b
1
e
1
+
m

j=2
b
j
e
j
and the two vectors b
1
e
1
and

m
j=2
b
j
e
j
are orthogonal. By Pythagoras’ theorem, the square of the length b of b is
b
2
= b
2
1
+
m

j=2
b
j
e
j

2
.
Pythagoras’ theorem can now be applied again to the last sum by singling out its first term b
2
e
2
, and so forth. In
conclusion,
b
2
=
m

j=1
b
2
j
.
This result extends Pythagoras’ theorem to m dimensions.
If we define the inner product of two m-dimensional vectors as follows:
b
T
c =
m

j=1
b
j
c
j
,
then
b
2
= b
T
b . (2.8)
Thus, the squared length of a vector is the inner product of the vector with itself. Here and elsewhere, vectors are
column vectors by default, and the symbol
T
makes them into row vectors.
12 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
Theorem 2.3.1
b
T
c = b c cos θ
where θ is the angle between b and c.
Proof. The law of cosines applied to the triangle with sides b, c, and b −c yields
b −c
2
= b
2
+c
2
−2b c cos θ
and from equation (2.8) we obtain
b
T
b + c
T
c −2b
T
c = b
T
b + c
T
c −2b c cos θ .
Canceling equal terms and dividing by -2 yields the desired result. ∆
Corollary 2.3.2 Two nonzero vectors b and c in R
m
are mutually orthogonal iff b
T
c = 0.
Proof. When θ = ±π/2, the previous theorem yields b
T
c = 0. ∆
Given two vectors b and c applied to the origin, the projection of b onto c is the vector from the origin to the point
p on the line through c that is nearest to the endpoint of b. See figure 2.2.
p
b
c
Figure 2.2: The vector from the origin to point p is the projection of b onto c. The line from the endpoint of b to p is
orthogonal to c.
Theorem 2.3.3 The projection of b onto c is the vector
p = P
c
b
where P
c
is the following square matrix:
P
c
=
cc
T
c
T
c
.
Proof. Since by definition point p is on the line through c, the projection vector p has the form p = ac, where
a is some real number. From elementary geometry, the line between p and the endpoint of b is shortest when it is
orthogonal to c:
c
T
(b −ac) = 0
2.4. ORTHOGONAL SUBSPACES AND THE RANK OF A MATRIX 13
which yields
a =
c
T
b
c
T
c
so that
p = ac = c a =
cc
T
c
T
c
b
as advertised. ∆
2.4 Orthogonal Subspaces and the Rank of a Matrix
Linear transformations map spaces into spaces. It is important to understand exactly what is being mapped into what
in order to determine whether a linear system has solutions, and if so how many. This section introduces the notion of
orthogonality between spaces, defines the null space and range of a matrix, and its rank. With these tools, we will be
able to characterize the solutions to a linear system in section 2.5. In the process, we also introduce a useful procedure
(Gram-Schmidt) for orthonormalizing a set of linearly independent vectors.
Two vector spaces A and B are said to be orthogonal to one another when every vector in A is orthogonal to every
vector in B. If vector space A is a subspace of R
m
for some m, then the orthogonal complement of A is the set of all
vectors in R
m
that are orthogonal to all the vectors in A.
Notice that complement and orthogonal complement are very different notions. For instance, the complement of
the xy plane in R
3
is all of R
3
except the xy plane, while the orthogonal complement of the xy plane is the z axis.
Theorem 2.4.1 Any basis a
1
, . . . , a
n
for a subspace A of R
m
can be extended into a basis for R
m
by adding m− n
vectors a
n+1
, . . . , a
m
.
Proof. If n = m we are done. If n < m, the given basis cannot generate all of R
m
, so there must be a vector, call
it a
n+1
, that is linearly independent of a
1
, . . . , a
n
. This argument can be repeated until the basis spans all of R
m
, that
is, until m = n. ∆
Theorem 2.4.2 (Gram-Schmidt) Given n vectors a
1
, . . . , a
n
, the following construction
r = 0
for j = 1 to n
a

j
= a
j

r
l=1
(q
T
l
a
j
)q
l
if a

j
= 0
r = r + 1
q
r
=
a

j
a

j

end
end
yields a set of orthonormal
3
vectors q
1
. . . , q
r
that span the same space as a
1
, . . . , a
n
.
Proof. We first prove by induction on r that the vectors q
r
are mutually orthonormal. If r = 1, there is little to
prove. The normalization in the above procedure ensures that q
1
has unit norm. Let us now assume that the procedure
3
Orthonormal means orthogonal and with unit norm.
14 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
above has been performed a number j −1 of times sufficient to find r −1 vectors q
1
, . . . , q
r−1
, and that these vectors
are orthonormal (the inductive assumption). Then for any i < r we have
q
T
i
a

j
= q
T
i
a
j

r−1

l=1
(q
T
l
a
j
)q
T
i
q
l
= 0
because the term q
T
i
a
j
cancels the i-th term (q
T
i
a
j
)q
T
i
q
i
of the sum (remember that q
T
i
q
i
= 1), and the inner products
q
T
i
q
l
are zero by the inductive assumption. Because of the explicit normalization step q
r
= a

j
/a

j
, the vector q
r
, if
computed, has unit norm, and because q
T
i
a

j
= 0, it follwos that q
r
is orthogonal to all its predecessors, q
T
i
q
r
= 0 for
i = 1, . . . , r −1.
Finally, we notice that the vectors q
j
span the same space as the a
j
s, because the former are linear combinations
of the latter, are orthonormal (and therefore independent), and equal in number to the number of linearly independent
vectors in a
1
, . . . , a
n
. ∆
Theorem 2.4.3 If A is a subspace of R
m
and A

is the orthogonal complement of A in R
m
, then
dim(A) + dim(A

) = m .
Proof. Let a
1
, . . . , a
n
be a basis for A. Extend this basis to a basis a
1
, . . . , a
m
for R
m
(theorem 2.4.1). Orthonor-
malize this basis by the Gram-Schmidt procedure (theorem 2.4.2) to obtain q
1
, . . . , q
m
. By construction, q
1
, . . . , q
n
span A. Because the new basis is orthonormal, all vectors generated by q
n+1
, . . . , q
m
are orthogonal to all vectors
generated by q
1
, . . . , q
n
, so there is a space of dimension at least m− n that is orthogonal to A. On the other hand,
the dimension of this orthogonal space cannot exceed m−n, because otherwise we would have more than m vectors
in a basis for R
m
. Thus, the dimension of the orthogonal space A

is exactly m−n, as promised. ∆
We can now start to talk about matrices in terms of the subspaces associated with them. The null space null(A) of
an m× n matrix A is the space of all n-dimensional vectors that are orthogonal to the rows of A. The range of A is
the space of all m-dimensional vectors that are generated by the columns of A. Thus, x ∈ null(A) iff Ax = 0, and
b ∈ range(A) iff Ax = b for some x.
From theorem 2.4.3, if null(A) has dimension h, then the space generated by the rows of A has dimension r =
n −h, that is, A has n −h linearly independent rows. It is not obvious that the space generated by the columns of A
has also dimension r = n −h. This is the point of the following theorem.
Theorem 2.4.4 The number r of linearly independent columns of any m× n matrix A is equal to the number of its
independent rows, and
r = n −h
where h = dim(null(A)).
Proof. We have already proven that the number of independent rows is n − h. Now we show that the number of
independent columns is also n −h, by constructing a basis for range(A).
Let v
1
, . . . , v
h
be a basis for null(A), and extend this basis (theorem 2.4.1) into a basis v
1
, . . . , v
n
for R
n
. Then
we can show that the n −h vectors Av
h+1
, . . . , Av
n
are a basis for the range of A.
First, these n − h vectors generate the range of A. In fact, given an arbitrary vector b ∈ range(A), there must be
a linear combination of the columns of A that is equal to b. In symbols, there is an n-tuple x such that Ax = b. The
n-tuple x itself, being an element of R
n
, must be some linear combination of v
1
, . . . , v
n
, our basis for R
n
:
x =
n

j=1
c
j
v
j
.
2.5. THE SOLUTIONS OF A LINEAR SYSTEM 15
Thus,
b = Ax = A
n

j=1
c
j
v
j
=
n

j=1
c
j
Av
j
=
n

j=h+1
c
j
Av
j
since v
1
, . . . , v
h
span null(A), so that Av
j
= 0 for j = 1, . . . , h. This proves that the n −h vectors Av
h+1
, . . . , Av
n
generate range(A).
Second, we prove that the n − h vectors Av
h+1
, . . . , Av
n
are linearly independent. Suppose, per absurdum, that
they are not. Then there exist numbers x
h+1
, . . . , x
n
, not all zero, such that
n

j=h+1
x
j
Av
j
= 0
so that
A
n

j=h+1
x
j
v
j
= 0 .
But then the vector

n
j=h+1
x
j
v
j
is in the null space of A. Since the vectors v
1
, . . . , v
h
are a basis for null(A), there
must exist coefficients x
1
, . . . , x
h
such that
n

j=h+1
x
j
v
j
=
h

j=1
x
j
v
j
,
in conflict with the assumption that the vectors v
1
, . . . , v
n
are linearly independent. ∆
Thanks to this theorem, we can define the rank of Ato be equivalently the number of linearly independent columns
or of linearly independent rows of A:
rank(A) = dim(range(A)) = n −dim(null(A)) .
2.5 The Solutions of a Linear System
Thanks to the results of the previous sections, we now have a complete picture of the four spaces associated with an
m×n matrix A of rank r and null-space dimension h:
range(A); dimension r = rank(A)
null(A); dimension h
range(A)

; dimension m−r
null(A)

; dimension r = n −h .
The space range(A)

is called the left nullspace of the matrix, and null(A)

is called the rowspace of A. A
frequently used synonym for “range” is column space. It should be obvious from the meaning of these spaces that
null(A)

= range(A
T
)
range(A)

= null(A
T
)
where A
T
is the transpose of A, defined as the matrix obtained by exchanging the rows of A with its columns.
Theorem 2.5.1 The matrix A transforms a vector x in its null space into the zero vector, and an arbitrary vector x
into a vector in range(A).
16 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
This allows characterizing the set of solutions to linear system as follows. Let
Ax = b
be an m×n system (m can be less than, equal to, or greater than n). Also, let
r = rank(A)
be the number of linearly independent rows or columns of A. Then,
b ∈ range(A) ⇒ no solutions
b ∈ range(A) ⇒ ∞
n−r
solutions
with the convention that ∞
0
= 1. Here, ∞
k
is the cardinality of a k-dimensional vector space.
In the first case above, there can be no linear combination of the columns (no x vector) that gives b, and the system
is said to be incompatible. In the second, compatible case, three possibilities occur, depending on the relative sizes of
r, m, n:
• When r = n = m, the system is invertible. This means that there is exactly one x that satisfies the system, since
the columns of A span all of R
n
. Notice that invertibility depends only on A, not on b.
• When r = n and m > n, the system is redundant. There are more equations than unknowns, but since b is in
the range of A there is a linear combination of the columns (a vector x) that produces b. In other words, the
equations are compatible, and exactly one solution exists.
4
• When r < n the system is underdetermined. This means that the null space is nontrivial (i.e., it has dimension
h > 0), and there is a space of dimension h = n −r of vectors x such that Ax = 0. Since b is assumed to be in
the range of A, there are solutions x to Ax = b, but then for any y ∈ null(A) also x + y is a solution:
Ax = b , Ay = 0 ⇒ A(x + y) = b
and this generates the ∞
h
= ∞
n−r
solutions mentioned above.
Notice that if r = n then n cannot possibly exceed m, so the first two cases exhaust the possibilities for r = n. Also,
r cannot exceed either m or n. All the cases are summarized in figure 2.3.
Of course, listing all possibilities does not provide an operational method for determining the type of linear system
for a given pair A, b. Gaussian elimination, and particularly its version called reduction to echelon form is such a
method, and is summarized in the next section.
2.6 Gaussian Elimination
Gaussian elimination is an important technique for solving linear systems. In addition to always yielding a solution,
no matter whether the system is invertible or not, it also allows determining the rank of a matrix.
Other solution techniques exist for linear systems. Most notably, iterative methods solve systems in a time that
depends on the accuracy required, while direct methods, like Gaussian elimination, are done in a finite amount of
time that can be bounded given only the size of a matrix. Which method to use depends on the size and structure
(e.g., sparsity) of the matrix, whether more information is required about the matrix of the system, and on numerical
considerations. More on this in chapter 3.
Consider the m×n system
Ax = b (2.9)
4
Notice that the technical meaning of “redundant” has a stronger meaning than “with more equations than unknowns.” The case r < n < mis
possible, has more equations (m) than unknowns (n), admits a solution if b ∈ range(A), but is called “underdetermined” because there are fewer
(r) independent equations than there are unknowns (see next item). Thus, “redundant” means “with exactly one solution and with more equations
than unknowns.”
2.6. GAUSSIAN ELIMINATION 17
yes no
yes no
yes no
underdetermined
redundant invertible
b in range(A)
r = n
m = n
incompatible
Figure 2.3: Types of linear systems.
which can be square or rectangular, invertible, incompatible, redundant, or underdetermined. In short, there are no
restrictions on the system. Gaussian elimination replaces the rows of this system by linear combinations of the rows
themselves until A is changed into a matrix U that is in the so-called echelon form. This means that
• Nonzero rows precede rows with all zeros. The first nonzero entry, if any, of a row, is called a pivot.
• Below each pivot is a column of zeros.
• Each pivot lies to the right of the pivot in the row above.
The same operations are applied to the rows of A and to those of b, which is transformed to a new vector c, so equality
is preserved and solving the final system yields the same solution as solving the original one.
Once the system is transformed into echelon form, we compute the solution x by backsubstitution, that is, by
solving the transformed system
Ux = c .
2.6.1 Reduction to Echelon Form
The matrix A is reduced to echelon form by a process in m − 1 steps. The first step is applied to U
(1)
= A and
c
(1)
= b. The k-th step is applied to rows k, . . . , m of U
(k)
and c
(k)
and produces U
(k+1)
and c
(k+1)
. The last step
produces U
(m)
= U and c
(m)
= c. Initially, the “pivot column index” p is set to one. Here is step k, where u
ij
denotes
entry i, j of U
(k)
:
Skip no-pivot columns If u
ip
is zero for every i = k, . . . , m, then increment p by 1. If p exceeds n stop.
5
Row exchange Now p ≤ n and u
ip
is nonzero for some k ≤ i ≤ m. Let l be one such value of i
6
. If l = k, exchange
rows l and k of U
(k)
and of c
(k)
.
Triangularization The new entry u
kp
is nonzero, and is called the pivot. For i = k + 1, . . . , m, subtract row k of
U
(k)
multiplied by u
ip
/u
kp
from row i of U
(k)
, and subtract entry k of c
(k)
multiplied by u
ip
/u
kp
from entry i
of c
(k)
. This zeros all the entries in the column below the pivot, and preserves the equality of left- and right-hand
side.
When this process is finished, U is in echelon form. In particular, if the matrix is square and if all columns have a
pivot, then U is upper-triangular.
5
“Stop” means that the entire algorithm is finished.
6
Different ways of selecting l here lead to different numerical properties of the algorithm. Selecting the largest entry in the column leads to
better round-off properties.
18 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
2.6.2 Backsubstitution
A system
Ux = c (2.10)
in echelon form is easily solved for x. To see this, we first solve the system symbolically, leaving undetermined vari-
ables specified by their name, and then transformthis solution procedure into one that can be more readily implemented
numerically.
Let r be the index of the last nonzero row of U. Since this is the number of independent rows of U, r is the rank
of U. It is also the rank of A, because A and U admit exactly the same solutions and are equal in size. If r < m, the
last m−r equations yield a subsystem of the following form:
_
¸
_
0
.
.
.
0
_
¸
_ =
_
¸
_
c
r+1
.
.
.
c
m
_
¸
_ .
Let us call this the residual subsystem. If on the other hand r = m (obviously r cannot exceed m), there is no residual
subsystem.
If there is a residual system (i.e., r < m) and some of c
r+1
, . . . , c
m
are nonzero, then the equations corresponding
to these nonzero entries are incompatible, because they are of the form 0 = c
i
with c
i
= 0. Since no vector x can
satisfy these equations, the linear system admits no solutions: it is incompatible.
Let us now assume that either there is no residual system, or if there is one it is compatible, that is, c
r+1
= . . . =
c
m
= 0. Then, solutions exist, and they can be determined by backsubstitution, that is, by solving the equations
starting from the last one and replacing the result in the equations higher up.
Backsubstitutions works as follows. First, remove the residual system, if any. We are left with an r ×n system. In
this system, call the variables corresponding to the r columns with pivots the basic variables, and call the other n −r
the free variables. Say that the pivot columns are j
1
, . . . , j
r
. Then symbolic backsubstitution consists of the following
sequence:
for i = r downto 1
x
j
i
=
1
u
ij
i
_
_
c
i

n

l=ji+1
u
il
x
l
_
_
end
This is called symbolic backsubstitution because no numerical values are assigned to free variables. Whenever they
appear in the expressions for the basic variables, free variables are specified by name rather than by value. The final
result is a solution with as many free parameters as there are free variables. Since any value given to the free variables
leaves the equality of system (2.10) satisfied, the presence of free variables leads to an infinity of solutions.
When solving a system in echelon form numerically, however, it is inconvenient to carry around nonnumeric
symbol names (the free variables). Here is an equivalent solution procedure that makes this unnecessary. The solution
obtained by backsubstitution is an affine function
7
of the free variables, and can therefore be written in the form
x = v
0
+x
j
1
v
1
+. . . +x
j
n−r
v
n−r
(2.11)
where the x
j
i
are the free variables. The vector v
0
is the solution when all free variables are zero, and can therefore be
obtained by replacing each free variable by zero during backsubstitution. Similarly, the vector v
i
for i = 1, . . . , n −r
can be obtained by solving the homogeneous system
Ux = 0
with x
ji
= 1 and all other free variables equal to zero. In conclusion, the general solution can be obtained by running
backsubstitution n−r +1 times, once for the nonhomogeneous system, and n−r times for the homogeneous system,
with suitable values of the free variables. This yields the solution in the form (2.11).
Notice that the vectors v
1
, . . . , v
n−r
form a basis for the null space of U, and therefore of A.
7
An affine function is a linear function plus a constant.
2.6. GAUSSIAN ELIMINATION 19
2.6.3 An Example
An example will clarify both the reduction to echelon form and backsubstitution. Consider the system
Ax = b
where
U
(1)
= A =
_
_
1 3 3 2
2 6 9 5
−1 −3 3 0
_
_
, c
(1)
= b =
_
_
1
5
5
_
_
.
Reduction to echelon form transforms A and b as follows. In the first step (k = 1), there are no no-pivot columns,
so the pivot column index p stays at 1. Throughout this example, we choose a trivial pivot selection rule: we pick the
first nonzero entry at or below row k in the pivot column. For k = 1, this means that u
(1)
11
= a
11
= 1 is the pivot. In
other words, no row exchange is necessary.
8
The triangularization step subtracts row 1 multiplied by 2/1 from row 2,
and subtracts row 1 multiplied by -1/1 from row 3. When applied to both U
(1)
and c
(1)
this yields
U
(2)
=
_
_
1 3 3 2
0 0 3 1
0 0 6 2
_
_
, c
(2)
=
_
_
1
3
6
_
_
.
Notice that now (k = 2) the entries u
(2)
ip
are zero for i = 2, 3, for both p = 1 and p = 2, so p is set to 3: the second
pivot column is column 3, and u
(2)
23
is nonzero, so no row exchange is necessary. In the triangularization step, row 2
multiplied by 6/3 is subtracted from row 3 for both U
(2)
and c
(2)
to yield
U = U
(3)
=
_
_
1 3 3 2
0 0 3 1
0 0 0 0
_
_
, c = c
(3)
=
_
_
1
3
0
_
_
.
There is one zero row in the left-hand side, and the rank of U and that of A is r = 2, the number of nonzero rows.
The residual system is 0 = 0 (compatible), and r < n = 4, so the system is underdetermined, with ∞
n−r
= ∞
2
solutions.
In symbolic backsubstitution, the residual subsystem is first deleted. This yields the reduced system
_
1 3 3 2
0 0 3 1
_
x =
_
1
3
_
(2.12)
The basic variables are x
1
and x
3
, corresponding to the columns with pivots. The other two variables, x
2
and
x
4
, are free. Backsubstitution applied first to row 2 and then to row 1 yields the following expressions for the pivot
variables:
x
3
=
1
u
23
(c
2
−u
24
x
4
) =
1
3
(3 −x
4
) = 1 −
1
3
x
4
x
1
=
1
u
11
(c
1
−u
12
x
2
−u
13
x
3
−u
14
x
4
) =
1
1
(1 −3x
2
−3x
3
−2x
4
)
= 1 −3x
2
−(3 −x
4
) −2x
4
= −2 −3x
2
−x
4
so the general solution is
x =
_
¸
¸
_
−2 −3x
2
−x
4
x
2
1 −
1
3
x
4
x
4
_
¸
¸
_
=
_
¸
¸
_
−2
0
1
0
_
¸
¸
_
+x
2
_
¸
¸
_
−3
1
0
0
_
¸
¸
_
+x
4
_
¸
¸
_
−1
0

1
3
1
_
¸
¸
_
.
8
Selecting the largest entry in the column at or below row k is a frequent choice, and this would have caused rows 1 and 2 to be switched.
20 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
This same solution can be found by the numerical backsubstitution method as follows. Solving the reduced system
(2.12) with x
2
= x
4
= 0 by numerical backsubstitution yields
x
3
=
1
3
(3 −1 · 0) = 1
x
1
=
1
1
(1 −3 · 0 −3 · 1 −2 · 0) = −2
so that
v
0
=
_
¸
¸
_
−2
0
1
0
_
¸
¸
_
.
Then v
1
is found by solving the nonzero part (first two rows) of Ux = 0 with x
2
= 1 and x
4
= 0 to obtain
x
3
=
1
3
(−1 · 0) = 0
x
1
=
1
1
(−3 · 1 −3 · 0 −2 · 0) = −3
so that
v
1
=
_
¸
¸
_
−3
1
0
0
_
¸
¸
_
.
Finally, solving the nonzero part of Ux = 0 with x
2
= 0 and x
4
= 1 leads to
x
3
=
1
3
(−1 · 1) = −
1
3
x
1
=
1
1
(−3 · 0 −3 ·
_

1
3
_
−2 · 1) = −1
so that
v
2
=
_
¸
¸
_
−1
0

1
3
1
_
¸
¸
_
and
x = v
0
+x
2
v
1
+x
4
v
2
=
_
¸
¸
_
−2
0
1
0
_
¸
¸
_
+x
2
_
¸
¸
_
−3
1
0
0
_
¸
¸
_
+x
4
_
¸
¸
_
−1
0

1
3
1
_
¸
¸
_
just as before.
As mentioned at the beginning of this section, Gaussian elimination is a direct method, in the sense that the answer
can be found in a number of steps that depends only on the size of the matrix A. In the next chapter, we study a different
2.6. GAUSSIAN ELIMINATION 21
method, based on the so-called the Singular Value Decomposition (SVD). This is an iterative method, meaning that an
exact solution usually requires an infinite number of steps, and the number of steps necessary to find an approximate
solution depends on the desired number of correct digits.
This state of affairs would seem to favor Gaussian elimination over the SVD. However, the latter yields a much
more complete answer, since it computes bases for all the four spaces mentioned above, as well as a set of quantities,
called the singular values, which provide great insight into the behavior of the linear transformation represented by
the matrix A. Singular values also allow defining a notion of approximate rank which is very useful in a large number
of applications. It also allows finding approximate solutions when the linear system in question is incompatible. In
addition, for reasons that will become apparent in the next chapter, the computation of the SVD is numerically well
behaved, much more so than Gaussian elimination. Finally, very efficient algorithms for the SVD exist. For instance,
on a regular workstation, one can compute several thousand SVDs of 5 × 5 matrices in one second. More generally,
the number of floating point operations necessary to compute the SVD of an m×n matrix is amn
2
+bn
3
where a, b
are small numbers that depend on the details of the algorithm.
22 CHAPTER 2. ALGEBRAIC LINEAR SYSTEMS
Chapter 3
The Singular Value Decomposition
In section 2, we saw that a matrix transforms vectors in its domain into vectors in its range (column space), and vectors
in its null space into the zero vector. No nonzero vector is mapped into the left null space, that is, into the orthogonal
complement of the range. In this section, we make this statement more specific by showing how unit vectors
1
in the
rowspace are transformed by matrices. This describes the action that a matrix has on the magnitudes of vectors as
well. To this end, we first need to introduce the notion of orthogonal matrices, and interpret them geometrically as
transformations between systems of orthonormal coordinates. We do this in section 3.1. Then, in section 3.2, we use
these new concepts to introduce the all-important concept of the Singular Value Decomposition (SVD). The chapter
concludes with some basic applications and examples.
3.1 Orthogonal Matrices
Let S be an n-dimensional subspace of R
m
(so that we necessarily have n ≤ m), and let v
1
, . . . , v
n
be an orthonormal
basis for S. Consider a point P in S. If the coordinates of P in R
m
are collected in an m-dimensional vector
p =
_
¸
_
p
1
.
.
.
p
m
_
¸
_ ,
and since P is in S, it must be possible to write p as a linear combination of the v
j
s. In other words, there must exist
coefficients
q =
_
¸
_
q
1
.
.
.
q
n
_
¸
_
such that
p = q
1
v
1
+. . . +q
n
v
n
= V q
where
V =
_
v
1
· · · v
n
¸
is an m×n matrix that collects the basis for S as its columns. Then for any i = 1, . . . , n we have
v
T
i
p = v
T
i
n

j=1
q
j
v
j
=
n

j=1
q
j
v
T
i
v
j
= q
i
,
since the v
j
are orthonormal. This is important, and may need emphasis:
1
Vectors with unit norm.
23
24 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
If
p =
n

j=1
q
j
v
j
and the vectors of the basis v
1
, . . . , v
n
are orthonormal, then the coefficients q
j
are the signed mag-
nitudes of the projections of p onto the basis vectors:
q
j
= v
T
j
p . (3.1)
In matrix form,
q = V
T
p . (3.2)
Also, we can collect the n
2
equations
v
T
i
v
j
=
_
1 if i = j
0 otherwise
into the following matrix equation:
V
T
V = I (3.3)
where I is the n ×n identity matrix. A matrix V that satisfies equation (3.3) is said to be orthogonal. Thus, a matrix
is orthogonal if its columns are orthonormal. Since the left inverse of a matrix V is defined as the matrix L such that
LV = I , (3.4)
comparison with equation (3.3) shows that the left inverse of an orthogonal matrix V exists, and is equal to the
transpose of V .
Of course, this argument requires V to be full rank, so that the solution L to equation (3.4) is unique. However, V
is certainly full rank, because it is made of orthonormal columns.
Notice that V R = I cannot possibly have a solution when m > n, because the m × m identity matrix has m
linearly independent
2
columns, while the columns of V R are linear combinations of the n columns of V , so V R can
have at most n linearly independent columns.
Of course, this result is still valid when V is m×m and has orthonormal columns, since equation (3.3) still holds.
However, for square, full-rank matrices (r = m = n), the distinction between left and right inverse vanishes. In fact,
suppose that there exist matrices L and R such that LV = I and V R = I. Then L = L(V R) = (LV )R = R, so the
left and the right inverse are the same. Thus, for square orthogonal matrices, V
T
is both the left and the right inverse:
V
T
V = V V
T
= I ,
and V
T
is then simply said to be the inverse of V :
V
T
= V
−1
.
Since the matrix V V
T
contains the inner products between the rows of V (just as V
T
V is formed by the inner
products of its columns), the argument above shows that the rows of a square orthogonal matrix are orthonormal as
well. We can summarize this discussion as follows:
Theorem 3.1.1 The left inverse of an orthogonal m×n matrix V with m ≥ n exists and is equal to the transpose of
V :
V
T
V = I .
In particular, if m = n, the matrix V
−1
= V
T
is also the right inverse of V :
V square ⇒ V
−1
V = V
T
V = V V
−1
= V V
T
= I .
2
Nay, orthonormal.
3.1. ORTHOGONAL MATRICES 25
Sometimes, when m = n, the geometric interpretation of equation (3.2) causes confusion, because two interpreta-
tions of it are possible. In the interpretation given above, the point P remains the same, and the underlying reference
frame is changed from the elementary vectors e
j
(that is, from the columns of I) to the vectors v
j
(that is, to the
columns of V ). Alternatively, equation (3.2) can be seen as a transformation, in a fixed reference system, of point P
with coordinates p into a different point Q with coordinates q. This, however, is relativity, and should not be surpris-
ing: If you spin clockwise on your feet, or if you stand still and the whole universe spins counterclockwise around
you, the result is the same.
3
Consistently with either of these geometric interpretations, we have the following result:
Theorem 3.1.2 The norm of a vector x is not changed by multiplication by an orthogonal matrix V :
V x = x .
Proof.
V x
2
= x
T
V
T
V x = x
T
x = x
2
.

We conclude this section with an obvious but useful consequence of orthogonality. In section 2.3 we defined the
projection p of a vector b onto another vector c as the point on the line through c that is closest to b. This notion of
projection can be extended from lines to vector spaces by the following definition: The projection p of a point b ∈ R
n
onto a subspace C is the point in C that is closest to b.
Also, for unit vectors c, the projection matrix is cc
T
(theorem 2.3.3), and the vector b − p is orthogonal to c. An
analogous result holds for subspace projection, as the following theorem shows.
Theorem 3.1.3 Let U be an orthogonal matrix. Then the matrix UU
T
projects any vector b onto range(U). Further-
more, the difference vector between b and its projection p onto range(U) is orthogonal to range(U):
U
T
(b −p) = 0 .
Proof. A point p in range(U) is a linear combination of the columns of U:
p = Ux
where x is the vector of coefficients (as many coefficients as there are columns in U). The squared distance between b
and p is
b −p
2
= (b −p)
T
(b −p) = b
T
b + p
T
p −2b
T
p = b
T
b + x
T
U
T
Ux −2b
T
Ux .
Because of orthogonality, U
T
U is the identity matrix, so
b −p
2
= b
T
b + x
T
x −2b
T
Ux .
The derivative of this squared distance with respect to x is the vector
2x −2U
T
b
3
At least geometrically. One solution may be more efficient than the other in other ways.
26 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
x
b
2
v
1
u σ
v
u
3
2
x
1
x
2
2
b
b
3
1
2
u σ
1 1
b
Figure 3.1: The matrix in equation (3.5) maps a circle on the plane into an ellipse in space. The two small boxes are
corresponding points.
which is zero iff
x = U
T
b ,
that is, when
p = Ux = UU
T
b
as promised.
For this value of p the difference vector b −p is orthogonal to range(U), in the sense that
U
T
(b −p) = U
T
(b −UU
T
b) = U
T
b −U
T
b = 0 .

3.2 The Singular Value Decomposition
In these notes, we have often used geometric intuition to introduce newconcepts, and we have then translated these into
algebraic statements. This approach is successful when geometry is less cumbersome than algebra, or when geometric
intuition provides a strong guiding element. The geometric picture underlying the Singular Value Decomposition is
crisp and useful, so we will use geometric intuition again. Here is the main intuition:
An m × n matrix A of rank r maps the r-dimensional unit hypersphere in rowspace(A) into an r-
dimensional hyperellipse in range(A).
This statement is stronger than saying that A maps rowspace(A) into range(A), because it also describes what
happens to the magnitudes of the vectors: a hypersphere is stretched or compressed into a hyperellipse, which is a
quadratic hypersurface that generalizes the two-dimensional notion of ellipse to an arbitrary number of dimensions. In
three dimensions, the hyperellipse is an ellipsoid, in one dimension it is a pair of points. In all cases, the hyperellipse
in question is centered at the origin.
For instance, the rank-2 matrix
A =
1

2
_
_

3

3
−3 3
1 1
_
_
(3.5)
transforms the unit circle on the plane into an ellipse embedded in three-dimensional space. Figure 3.1 shows the map
b = Ax .
3.2. THE SINGULAR VALUE DECOMPOSITION 27
Two diametrically opposite points on the unit circle are mapped into the two endpoints of the major axis of the
ellipse, and two other diametrically opposite points on the unit circle are mapped into the two endpoints of the minor
axis of the ellipse. The lines through these two pairs of points on the unit circle are always orthogonal. This result can
be generalized to any m×n matrix.
Simple and fundamental as this geometric fact may be, its proof by geometric means is cumbersome. Instead, we
will prove it algebraically by first introducing the existence of the SVD and then using the latter to prove that matrices
map hyperspheres into hyperellipses.
Theorem 3.2.1 If A is a real m×n matrix then there exist orthogonal matrices
U =
_
u
1
· · · u
m
¸
∈ R
m×m
V =
_
v
1
· · · v
n
¸
∈ R
n×n
such that
U
T
AV = Σ = diag(σ
1
, . . . , σ
p
) ∈ R
m×n
where p = min(m, n) and σ
1
≥ . . . ≥ σ
p
≥ 0. Equivalently,
A = UΣV
T
.
Proof. Let x and y be unit vectors in R
n
and R
m
, respectively, and consider the bilinear form
z = y
T
Ax .
The set
S = {x, y | x ∈ R
n
, y ∈ R
m
, x = y = 1}
is compact, so that the scalar function z(x, y) must achieve a maximum value on S, possibly at more than one point
4
.
Let u
1
, v
1
be two unit vectors in R
m
and R
n
respectively where this maximum is achieved, and let σ
1
be the corre-
sponding value of z:
max
x=y=1
y
T
Ax = u
T
1
Av
1
= σ
1
.
It is easy to see that u
1
is parallel to the vector Av
1
. If this were not the case, their inner product u
T
1
Av
1
could
be increased by rotating u
1
towards the direction of Av
1
, thereby contradicting the fact that u
T
1
Av
1
is a maximum.
Similarly, by noticing that
u
T
1
Av
1
= v
T
1
A
T
u
1
and repeating the argument above, we see that v
1
is parallel to A
T
u
1
.
By theorems 2.4.1 and 2.4.2, u
1
and v
1
can be extended into orthonormal bases for R
m
and R
n
, respectively.
Collect these orthonormal basis vectors into orthogonal matrices U
1
and V
1
. Then
U
T
1
AV
1
= S
1
=
_
σ
1
0
T
0 A
1
_
.
In fact, the first column of AV
1
is Av
1
= σ
1
u
1
, so the first entry of U
T
1
AV
1
is u
T
1
σ
1
u
1
= σ
1
, and its other entries
are u
T
j
Av
1
= 0 because Av
1
is parallel to u
1
and therefore orthogonal, by construction, to u
2
, . . . , u
m
. A similar
argument shows that the entries after the first in the first row of S
1
are zero: the row vector u
T
1
A is parallel to v
T
1
, and
therefore orthogonal to v
2
, . . . , v
n
, so that u
T
1
Av
2
= . . . = u
T
1
Av
n
= 0.
The matrix A
1
has one fewer row and column than A. We can repeat the same construction on A
1
and write
U
T
2
A
1
V
2
= S
2
=
_
σ
2
0
T
0 A
2
_
4
Actually, at least at two points: if u
T
1
Av
1
is a maximum, so is (−u
1
)
T
A(−v
1
).
28 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
so that
_
1 0
T
0 U
T
2
_
U
T
1
AV
1
_
1 0
T
0 V
2
_
=
_
_
σ
1
0 0
T
0 σ
2
0
T
0 0 A
2
_
_
.
This procedure can be repeated until A
k
vanishes (zero rows or zero columns) to obtain
U
T
AV = Σ
where U
T
and V are orthogonal matrices obtained by multiplying together all the orthogonal matrices used in the
procedure, and
Σ = diag(σ
1
, . . . , σ
p
) .
Since matrices U and V are orthogonal, we can premultiply the matrix product in the theorem by U and postmultiply
it by V
T
to obtain
A = UΣV
T
,
which is the desired result.
It only remains to show that the elements on the diagonal of Σ are nonnegative and arranged in nonincreasing
order. To see that σ
1
≥ . . . ≥ σ
p
(where p = min(m, n)), we can observe that the successive maximization problems
that yield σ
1
, . . . , σ
p
are performed on a sequence of sets each of which contains the next. To show this, we just need
to show that σ
2
≤ σ
1
, and induction will do the rest. We have
σ
2
= max

ˆ
x=
ˆ
y=1
ˆy
T
A
1
ˆx = max

ˆ
x=
ˆ
y=1
_
0 ˆy
¸
T
S
1
_
0
ˆx
_
= max

ˆ
x=
ˆ
y=1
_
0 ˆy
¸
T
U
T
1
AV
1
_
0
ˆx
_
= max
x = y = 1
x
T
v
1
= y
T
u
1
= 0
y
T
Ax ≤ σ
1
.
To explain the last equality above, consider the vectors
x = V
1
_
0
ˆx
_
and y = U
1
_
0
ˆy
_
.
The vector x is equal to the unit vector [0 ˆx]
T
transformed by the orthogonal matrix V
1
, and is therefore itself a unit
vector. In addition, it is a linear combination of v
2
, . . . , v
n
, and is therefore orthogonal to v
1
. A similar argument
shows that y is a unit vector orthogonal to u
1
. Because x and y thus defined belong to subsets (actually sub-spheres)
of the unit spheres in R
n
and R
m
, we conclude that σ
2
≤ σ
1
.
The σ
i
are nonnegative because all these maximizations are performed on unit hyper-spheres. The σ
i
s are maxima
of the function z(x, y) which always assumes both positive and negative values on any hyper-sphere: If z(x, y) is
negative, then z(−x, y) is positive, and if x is on a hyper-sphere, so is −x. ∆
We can now review the geometric picture in figure 3.1 in light of the singular value decomposition. In the process,
we introduce some nomenclature for the three matrices in the SVD. Consider the map in figure 3.1, represented by
equation (3.5), and imagine transforming point x (the small box at x on the unit circle) into its corresponding point
b = Ax (the small box on the ellipse). This transformation can be achieved in three steps (see figure 3.2):
1. Write x in the frame of reference of the two vectors v
1
, v
2
on the unit circle that map into the major axes of the
ellipse. There are a few ways to do this, because axis endpoints come in pairs. Just pick one way, but order
v
1
, v
2
so they map into the major and the minor axis, in this order. Let us call v
1
, v
2
the two right singular
vectors of A. The corresponding axis unit vectors u
1
, u
2
on the ellipse are called left singular vectors. If we
define
V =
_
v
1
v
2
¸
,
3.2. THE SINGULAR VALUE DECOMPOSITION 29
2
x
1
x
v
2
v
1
2 2
2
v’
1
v’
2
y
y
1
u
3
y
3
u σ
2 2
u σ
1 1
σ
2 2
u’
σ
1 1
u’
x
ξ
ξ
1
ξ
η
η
1
y
η
Figure 3.2: Decomposition of the mapping in figure 3.1.
the new coordinates ξ of x become
ξ = V
T
x
because V is orthogonal.
2. Transform ξ into its image on a “straight” version of the final ellipse. “Straight” here means that the axes of the
ellipse are aligned with the y
1
, y
2
axes. Otherwise, the “straight” ellipse has the same shape as the ellipse in
figure 3.1. If the lengths of the half-axes of the ellipse are σ
1
, σ
2
(major axis first), the transformed vector η has
coordinates
η = Σξ
where
Σ =
_
_
σ
1
0
0 σ
2
0 0
_
_
is a diagonal matrix. The real, nonnegative numbers σ
1
, σ
2
are called the singular values of A.
3. Rotate the reference frame in R
m
= R
3
so that the “straight” ellipse becomes the ellipse in figure 3.1. This
rotation brings η along, and maps it to b. The components of η are the signed magnitudes of the projections of
b along the unit vectors u
1
, u
2
, u
3
that identify the axes of the ellipse and the normal to the plane of the ellipse,
so
b = Uη
where the orthogonal matrix
U =
_
u
1
u
2
u
3
¸
collects the left singular vectors of A.
30 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
We can concatenate these three transformations to obtain
b = UΣV
T
x
or
A = UΣV
T
since this construction works for any point x on the unit circle. This is the SVD of A.
The singular value decomposition is “almost unique”. There are two sources of ambiguity. The first is in the
orientation of the singular vectors. One can flip any right singular vector, provided that the corresponding left singular
vector is flipped as well, and still obtain a valid SVD. Singular vectors must be flipped in pairs (a left vector and its
corresponding right vector) because the singular values are required to be nonnegative. This is a trivial ambiguity. If
desired, it can be removed by imposing, for instance, that the first nonzero entry of every left singular value be positive.
The second source of ambiguity is deeper. If the matrix A maps a hypersphere into another hypersphere, the axes
of the latter are not defined. For instance, the identity matrix has an infinity of SVDs, all of the form
I = UIU
T
where U is any orthogonal matrix of suitable size. More generally, whenever two or more singular values coincide,
the subspaces identified by the corresponding left and right singular vectors are unique, but any orthonormal basis can
be chosen within, say, the right subspace and yield, together with the corresponding left singular vectors, a valid SVD.
Except for these ambiguities, the SVD is unique.
Even in the general case, the singular values of a matrix A are the lengths of the semi-axes of the hyperellipse E
defined by
E = {Ax : x = 1} .
The SVD reveals a great deal about the structure of a matrix. If we define r by
σ
1
≥ . . . ≥ σ
r
> σ
r+1
= . . . = 0 ,
that is, if σ
r
is the smallest nonzero singular value of A, then
rank(A) = r
null(A) = span{v
r+1
, . . . , v
n
}
range(A) = span{u
1
, . . . , u
r
} .
The sizes of the matrices in the SVD are as follows: U is m× m, Σ is m× n, and V is n × n. Thus, Σ has the
same shape and size as A, while U and V are square. However, if m > n, the bottom (m−n) ×n block of Σ is zero,
so that the last m− n columns of U are multiplied by zero. Similarly, if m < n, the rightmost m× (n − m) block
of Σ is zero, and this multiplies the last n −m rows of V . This suggests a “small,” equivalent version of the SVD. If
p = min(m, n), we can define U
p
= U(:, 1 : p), Σ
p
= Σ(1 : p, 1 : p), and V
p
= V (:, 1 : p), and write
A = U
p
Σ
p
V
T
p
where U
p
is m×p, Σ
p
is p ×p, and V
p
is n ×p.
Moreover, if p−r singular values are zero, we can let U
r
= U(:, 1 : r), Σ
r
= Σ(1 : r, 1 : r), and V
r
= V (:, 1 : r),
then we have
A = U
r
Σ
r
V
T
r
=
r

i=1
σ
i
u
i
v
T
i
,
which is an even smaller, minimal, SVD.
Finally, both the 2-norm and the Frobenius norm
A
F
=
¸
¸
¸
_
m

i=1
n

j=1
|a
ij
|
2
3.3. THE PSEUDOINVERSE 31
and
A
2
= sup
x=0
Ax
x
are neatly characterized in terms of the SVD:
A
2
F
= σ
2
1
+. . . +σ
2
p
A
2
= σ
1
.
In the next few sections we introduce fundamental results and applications that testify to the importance of the
SVD.
3.3 The Pseudoinverse
One of the most important applications of the SVD is the solution of linear systems in the least squares sense. A linear
system of the form
Ax = b (3.6)
arising froma real-life application may or may not admit a solution, that is, a vector x that satisfies this equation exactly.
Often more measurements are available than strictly necessary, because measurements are unreliable. This leads to
more equations than unknowns (the number m of rows in A is greater than the number n of columns), and equations
are often mutually incompatible because they come from inexact measurements (incompatible linear systems were
defined in chapter 2). Even when m ≤ n the equations can be incompatible, because of errors in the measurements
that produce the entries of A. In these cases, it makes more sense to find a vector x that minimizes the norm
Ax −b
of the residual vector
r = Ax −b .
where the double bars henceforth refer to the Euclidean norm. Thus, x cannot exactly satisfy any of the m equations
in the system, but it tries to satisfy all of them as closely as possible, as measured by the sum of the squares of the
discrepancies between left- and right-hand sides of the equations.
In other circumstances, not enough measurements are available. Then, the linear system (3.6) is underdetermined,
in the sense that it has fewer independent equations than unknowns (its rank r is less than n, see again chapter 2).
Incompatibility and underdeterminacy can occur together: the system admits no solution, and the least-squares
solution is not unique. For instance, the system
x
1
+x
2
= 1
x
1
+x
2
= 3
x
3
= 2
has three unknowns, but rank 2, and its first two equations are incompatible: x
1
+ x
2
cannot be equal to both 1 and
3. A least-squares solution turns out to be x = [1 1 2]
T
with residual r = Ax − b = [1 − 1 0], which has norm

2
(admittedly, this is a rather high residual, but this is the best we can do for this problem, in the least-squares sense).
However, any other vector of the form
x

=
_
_
1
1
2
_
_

_
_
−1
1
0
_
_
is as good as x. For instance, x

= [0 2 2], obtained for α = 1, yields exactly the same residual as x (check this).
In summary, an exact solution to the system (3.6) may not exist, or may not be unique, as we learned in chapter 2.
An approximate solution, in the least-squares sense, always exists, but may fail to be unique.
32 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
If there are several least-squares solutions, all equally good (or bad), then one of them turns out to be shorter than
all the others, that is, its norm x is smallest. One can therefore redefine what it means to “solve” a linear system so
that there is always exactly one solution. This minimum norm solution is the subject of the following theorem, which
both proves uniqueness and provides a recipe for the computation of the solution.
Theorem 3.3.1 The minimum-norm least squares solution to a linear system Ax = b, that is, the shortest vector x
that achieves the
min
x
Ax −b ,
is unique, and is given by
ˆx = V Σ

U
T
b (3.7)
where
Σ

=
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
1/σ
1
0 · · · 0
.
.
.
1/σ
r
.
.
.
.
.
.
0
.
.
.
0 0 · · · 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
is an n ×m diagonal matrix.
The matrix
A

= V Σ

U
T
is called the pseudoinverse of A.
Proof. The minimum-norm Least Squares solution to
Ax = b
is the shortest vector x that minimizes
Ax −b
that is,
UΣV
T
x −b .
This can be written as
U(ΣV
T
x −U
T
b) (3.8)
because U is an orthogonal matrix, UU
T
= I. But orthogonal matrices do not change the norm of vectors they are
applied to (theorem 3.1.2), so that the last expression above equals
ΣV
T
x −U
T
b
or, with y = V
T
x and c = U
T
b,
Σy −c .
In order to find the solution to this minimization problem, let us spell out the last expression. We want to minimize the
norm of the following vector:
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
σ
1
0 · · · 0
0
.
.
.
· · · 0
σ
r
.
.
. 0
.
.
.
.
.
.
0 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
y
1
.
.
.
y
r
y
r+1
.
.
.
y
n
_
¸
¸
¸
¸
¸
¸
¸
¸
_

_
¸
¸
¸
¸
¸
¸
¸
¸
_
c
1
.
.
.
c
r
c
r+1
.
.
.
c
m
_
¸
¸
¸
¸
¸
¸
¸
¸
_
.
3.4. LEAST-SQUARES SOLUTION OF A HOMOGENEOUS LINEAR SYSTEMS 33
The last m−r differences are of the form
0 −
_
¸
_
c
r+1
.
.
.
c
m
_
¸
_
and do not depend on the unknown y. In other words, there is nothing we can do about those differences: if some or
all the c
i
for i = r +1, . . . , m are nonzero, we will not be able to zero these differences, and each of them contributes
a residual |c
i
| to the solution. In each of the first r differences, on the other hand, the last n − r components of y are
multiplied by zeros, so they have no effect on the solution. Thus, there is freedom in their choice. Since we look for
the minimum-norm solution, that is, for the shortest vector x, we also want the shortest y, because x and y are related
by an orthogonal transformation. We therefore set y
r+1
= . . . = y
n
= 0. In summary, the desired y has the following
components:
y
i
=
c
i
σ
i
for i = 1, . . . , r
y
i
= 0 for i = r + 1, . . . , n .
When written as a function of the vector c, this is
y = Σ
+
c .
Notice that there is no other choice for y, which is therefore unique: minimum residual forces the choice of y
1
, . . . , y
r
,
and minimum-norm solution forces the other entries of y. Thus, the minimum-norm, least-squares solution to the
original system is the unique vector
ˆx = V y = V Σ
+
c = V Σ
+
U
T
b
as promised. The residual, that is, the norm of Ax − b when x is the solution vector, is the norm of Σy − c, since
this vector is related to Ax −b by an orthogonal transformation (see equation (3.8)). In conclusion, the square of the
residual is
Ax −b
2
= Σy −c
2
=
m

i=r+1
c
2
i
=
m

i=r+1
(u
T
i
b)
2
which is the projection of the right-hand side vector b onto the complement of the range of A. ∆
3.4 Least-Squares Solution of a Homogeneous Linear Systems
Theorem 3.3.1 works regardless of the value of the right-hand side vector b. When b = 0, that is, when the system is
homogeneous, the solution is trivial: the minimum-norm solution to
Ax = 0 (3.9)
is
x = 0 ,
which happens to be an exact solution. Of course it is not necessarily the only one (any vector in the null space of A
is also a solution, by definition), but it is obviously the one with the smallest norm.
Thus, x = 0 is the minimum-norm solution to any homogeneous linear system. Although correct, this solution is
not too interesting. In many applications, what is desired is a nonzero vector x that satisfies the system (3.9) as well
as possible. Without any constraints on x, we would fall back to x = 0 again. For homogeneous linear systems, the
meaning of a least-squares solution is therefore usually modified, once more, by imposing the constraint
x = 1
34 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
on the solution. Unfortunately, the resulting constrained minimization problem does not necessarily admit a unique
solution. The following theorem provides a recipe for finding this solution, and shows that there is in general a whole
hypersphere of solutions.
Theorem 3.4.1 Let
A = UΣV
T
be the singular value decomposition of A. Furthermore, let v
n−k+1
, . . . , v
n
be the k columns of V whose correspond-
ing singular values are equal to the last singular value σ
n
, that is, let k be the largest integer such that
σ
n−k+1
= . . . = σ
n
.
Then, all vectors of the form
x = α
1
v
n−k+1
+. . . +α
k
v
n
(3.10)
with
α
2
1
+. . . +α
2
k
= 1 (3.11)
are unit-norm least squares solutions to the homogeneous linear system
Ax = 0,
that is, they achieve the
min
x=1
Ax .
Note: when σ
n
is greater than zero the most common case is k = 1, since it is very unlikely that different singular
values have exactly the same numerical value. When A is rank deficient, on the other case, it may often have more
than one singular value equal to zero. In any event, if k = 1, then the minimum-norm solution is unique, x = v
n
. If
k > 1, the theorem above shows how to express all solutions as a linear combination of the last k columns of V .
Proof. The reasoning is very similar to that for the previous theorem. The unit-norm Least Squares solution to
Ax = 0
is the vector x with x = 1 that minimizes
Ax
that is,
UΣV
T
x .
Since orthogonal matrices do not change the norm of vectors they are applied to (theorem 3.1.2), this norm is the same
as
ΣV
T
x
or, with y = V
T
x,
Σy .
Since V is orthogonal, x = 1 translates to y = 1. We thus look for the unit-norm vector y that minimizes the
norm (squared) of Σy, that is,
σ
2
1
y
2
1
+. . . +σ
2
n
y
2
n
.
This is obviously achieved by concentrating all the (unit) mass of y where the σs are smallest, that is by letting
y
1
= . . . = y
n−k
= 0. (3.12)
From y = V
T
x we obtain x = V y = y
1
v
1
+. . . +y
n
v
n
, so that equation (3.12) is equivalent to equation (3.10) with
α
1
= y
n−k+1
, . . . , α
k
= y
n
, and the unit-norm constraint on y yields equation (3.11). ∆
Section 3.5 shows a sample use of theorem 3.4.1.
3.5. SVD LINE FITTING 35
3.5 SVD Line Fitting
The Singular Value Decomposition of a matrix yields a simple method for fitting a line to a set of points on the plane.
3.5.1 Fitting a Line to a Set of Points
Let p
i
= (x
i
, y
i
)
T
be a set of m ≥ 2 points on the plane, and let
ax +by −c = 0
be the equation of a line. If the lefthand side of this equation is multiplied by a nonzero constant, the line does not
change. Thus, we can assume without loss of generality that
n = a
2
+b
2
= 1 , (3.13)
where the unit vector n = (a, b)
T
, orthogonal to the line, is called the line normal.
The distance from the line to the origin is |c| (see figure 3.3), and the distance between the line n and a point p
i
is
equal to
d
i
= |ax
i
+by
i
−c| = |p
T
i
n −c| . (3.14)
p
i
a
b
|c|
Figure 3.3: The distance between point p
i
= (x
i
, y
i
)
T
and line ax +by −c = 0 is |ax
i
+by
i
−c|.
The best-fit line minimizes the sum of the squared distances. Thus, if we let d = (d
1
, . . . , d
m
) and P =
(p
1
. . . , p
m
)
T
, the best-fit line achieves the
min
n=1
d
2
= min
n=1
Pn −c1
2
. (3.15)
In equation (3.15), 1 is a vector of m ones.
3.5.2 The Best Line Fit
Since the third line parameter c does not appear in the constraint (3.13), at the minimum (3.15) we must have
∂d
2
∂c
= 0 . (3.16)
If we define the centroid p of all the points p
i
as
p =
1
m
P
T
1 ,
36 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
equation (3.16) yields
∂d
2
∂c
=

∂c
_
n
T
P
T
−c1
T
_
(Pn −1c)
=

∂c
_
n
T
P
T
Pn +c
2
1
T
1 −2n
T
P
T
c1
_
= 2
_
mc −n
T
P
T
1
_
= 0
from which we obtain
c =
1
m
n
T
P
T
1 ,
that is,
c = p
T
n .
By replacing this expression into equation (3.15), we obtain
min
n=1
d
2
= min
n=1
Pn −1p
T
n
2
= min
n=1
Qn
2
,
where Q = P − 1p
T
collects the centered coordinates of the m points. We can solve this constrained minimization
problem by theorem 3.4.1. Equivalently, and in order to emphasize the geometric meaning of signular values and
vectors, we can recall that if n is on a circle, the shortest vector of the form Qn is obtained when n is the right singular
vector v
2
corresponding to the smaller σ
2
of the two singular values of Q. Furthermore, since Qv
2
has norm σ
2
, the
residue is
min
n=1
d = σ
2
and more specifically the distances d
i
are given by
d = σ
2
u
2
where u
2
is the left singular vector corresponding to σ
2
. In fact, when n = v
2
, the SVD
Q = UΣV
T
=
2

i=1
σ
i
u
i
v
T
i
yields
Qn = Qv
2
=
2

i=1
σ
i
u
i
v
T
i
v
2
= σ
2
u
2
because v
1
and v
2
are orthonormal vectors.
To summarize, to fit a line (a, b, c) to a set of m points p
i
collected in the m × 2 matrix P = (p
1
. . . , p
m
)
T
,
proceed as follows:
1. compute the centroid of the points (1 is a vector of m ones):
p =
1
m
P
T
1
2. form the matrix of centered coordinates:
Q = P −1p
T
3. compute the SVD of Q:
Q = UΣV
T
3.5. SVD LINE FITTING 37
4. the line normal is the second column of the 2 ×2 matrix V :
n = (a, b)
T
= v
2
,
5. the third coefficient of the line is
c = p
T
n
6. the residue of the fit is
min
n=1
d = σ
2
The following matlab code implements the line fitting method.
function [l, residue] = linefit(P)
% check input matrix sizes
[m n] = size(P);
if n ˜= 2, error(’matrix P must be m x 2’), end
if m < 2, error(’Need at least two points’), end
one = ones(m, 1);
% centroid of all the points
p = (P’ * one) / m;
% matrix of centered coordinates
Q = P - one * p’;
[U Sigma V] = svd(Q);
% the line normal is the second column of V
n = V(:, 2);
% assemble the three line coefficients into a column vector
l = [n ; p’ * n];
% the smallest singular value of Q
% measures the residual fitting error
residue = Sigma(2, 2);
A useful exercise is to think how this procedure, or something close to it, can be adapted to fit a set of data points
in R
m
with an affine subspace of given dimension n. An affine subspace is a linear subspace plus a point, just like an
arbitrary line is a line through the origin plus a point. Here “plus” means the following. Let L be a linear space. Then
an affine space has the form
A = p +L = {a | a = p + l and l ∈ L} .
Hint: minimizing the distance between a point and a subspace is equivalent to maximizing the norm of the projection
of the point onto the subspace. The fitting problem (including fitting a line to a set of points) can be cast either as a
maximization or a minimization problem.
38 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION
Chapter 4
Function Optimization
There are three main reasons why most problems in robotics, vision, and arguably every other science or endeavor
take on the form of optimization problems. One is that the desired goal may not be achievable, and so we try to get as
close as possible to it. The second reason is that there may be more ways to achieve the goal, and so we can choose
one by assigning a quality to all the solutions and selecting the best one. The third reason is that we may not know
how to solve the system of equations f(x) = 0, so instead we minimize the norm f(x), which is a scalar function of
the unknown vector x.
We have encountered the first two situations when talking about linear systems. The case in which a linear system
admits exactly one exact solution is simple but rare. More often, the system at hand is either incompatible (some say
overconstrained) or, at the opposite end, underdetermined. In fact, some problems are both, in a sense. While these
problems admit no exact solution, they often admit a multitude of approximate solutions. In addition, many problems
lead to nonlinear equations.
Consider, for instance, the problem of Structure From Motion (SFM) in computer vision. Nonlinear equations
describe how points in the world project onto the images taken by cameras at given positions in space. Structure from
motion goes the other way around, and attempts to solve these equations: image points are given, and one wants to
determine where the points in the world and the cameras are. Because image points come from noisy measurements,
they are not exact, and the resulting system is usually incompatible. SFM is then cast as an optimization problem.
On the other hand, the exact system (the one with perfect coefficients) is often close to being underdetermined. For
instance, the images may be insufficient to recover a certain shape under a certain motion. Then, an additional criterion
must be added to define what a “good” solution is. In these cases, the noisy system admits no exact solutions, but has
many approximate ones.
The term “optimization” is meant to subsume both minimization and maximization. However, maximizing the
scalar function f(x) is the same as minimizing −f(x), so we consider optimization and minimization to be essentially
synonyms. Usually, one is after global minima. However, global minima are hard to find, since they involve a universal
quantifier: x

is a global minimum of f if for every other x we have f(x) ≥ f(x

). Global minization techniques
like simulated annealing have been proposed, but their convergence properties depend very strongly on the problem at
hand. In this chapter, we consider local minimization: we pick a starting point x
0
, and we descend in the landscape of
f(x) until we cannot go down any further. The bottom of the valley is a local minimum.
Local minimization is appropriate if we know how to pick an x
0
that is close to x

. This occurs frequently in
feedback systems. In these systems, we start at a local (or even a global) minimum. The system then evolves and
escapes from the minimum. As soon as this occurs, a control signal is generated to bring the system back to the
minimum. Because of this immediate reaction, the old minimum can often be used as a starting point x
0
when looking
for the new minimum, that is, when computing the required control signal. More formally, we reach the correct
minimum x

as long as the initial point x
0
is in the basin of attraction of x

, defined as the largest neighborhood of x

in which f(x) is convex.
Good references for the discussion in this chapter are Matrix Computations, Practical Optimization, and Numerical
Recipes in C, all of which are listed with full citations in section 1.4.
39
40 CHAPTER 4. FUNCTION OPTIMIZATION
4.1 Local Minimization and Steepest Descent
Suppose that we want to find a local minimum for the scalar function f of the vector variable x, starting from an initial
point x
0
. Picking an appropriate x
0
is crucial, but also very problem-dependent. We start from x
0
, and we go downhill.
At every step of the way, we must make the following decisions:
• Whether to stop.
• In what direction to proceed.
• How long a step to take.
In fact, most minimization algorithms have the following structure:
k = 0
while x
k
is not a minimum
compute step direction p
k
with p
k
= 1
compute step size α
k
x
k+1
= x
k

k
p
k
k = k + 1
end.
Different algorithms differ in how each of these instructions is performed.
It is intuitively clear that the choice of the step size α
k
is important. Too small a step leads to slow convergence,
or even to lack of convergence altogether. Too large a step causes overshooting, that is, leaping past the solution. The
most disastrous consequence of this is that we may leave the basin of attraction, or that we oscillate back and forth
with increasing amplitudes, leading to instability. Even when oscillations decrease, they can slow down convergence
considerably.
What is less obvious is that the best direction of descent is not necessarily, and in fact is quite rarely, the direction
of steepest descent, as we now show. Consider a simple but important case,
f(x) = c + a
T
x +
1
2
x
T
Qx (4.1)
where Q is a symmetric, positive definite matrix. Positive definite means that for every nonzero x the quantity x
T
Qx
is positive. In this case, the graph of f(x) −c is a plane a
T
x plus a paraboloid.
Of course, if f were this simple, no descent methods would be necessary. In fact the minimum of f can be found
by setting its gradient to zero:
∂f
∂x
= a +Qx = 0
so that the minimum x

is the solution to the linear system
Qx = −a . (4.2)
Since Q is positive definite, it is also invertible (why?), and the solution x

is unique. However, understanding the
behavior of minimization algorithms in this simple case is crucial in order to establish the convergence properties of
these algorithms for more general functions. In fact, all smooth functions can be approximated by paraboloids in a
sufficiently small neighborhood of any point.
Let us therefore assume that we minimize f as given in equation (4.1), and that at every step we choose the
direction of steepest descent. In order to simplify the mathematics, we observe that if we let
˜ e(x) =
1
2
(x −x

)
T
Q(x −x

)
then we have
˜ e(x) = f(x) −c +
1
2
x
∗T
Qx

= f(x) −f(x

) (4.3)
4.1. LOCAL MINIMIZATION AND STEEPEST DESCENT 41
so that ˜ e and f differ only by a constant. In fact,
˜ e(x) =
1
2
(x
T
Qx + x
∗T
Qx

−2x
T
Qx

) =
1
2
x
T
Qx + a
T
x +
1
2
x
∗T
Qx

= f(x) −c +
1
2
x
∗T
Qx

and from equation (4.2) we obtain
f(x

) = c + a
T
x

+
1
2
x
∗T
Qx

= c −x
∗T
Qx

+
1
2
x
∗T
Qx

= c −
1
2
x
∗T
Qx

.
The result (4.3),
˜ e(x) = f(x) −f(x

) ,
is rather interesting in itself. It says that adding a linear term a
T
x (and a constant c) to a paraboloid
1
2
x
T
Qx merely
shifts the bottom of the paraboloid, both in position (x

rather than 0) and value (c −
1
2
x
∗T
Qx

rather than zero).
Adding the linear term does not “warp” or “tilt” the shape of the paraboloid in any way.
Since ˜ e is simpler, we consider that we are minimizing ˜ e rather than f. In addition, we can let
y = x −x

,
that is, we can shift the origin of the domain to x

, and study the function
e(y) =
1
2
y
T
Qy
instead of f or ˜ e, without loss of generality. We will transform everything back to f and x once we are done. Of
course, by construction, the new minimum is at
y

= 0
where e reaches a value of zero:
e(y

) = e(0) = 0 .
However, we let our steepest descent algorithm find this minimum by starting from the initial point
y
0
= x
0
−x

.
At every iteration k, the algorithm chooses the direction of steepest descent, which is in the direction
p
k
= −
g
k
g
k

opposite to the gradient of e evaluated at y
k
:
g
k
= g(y
k
) =
∂e
∂y
¸
¸
¸
¸
y=y
k
= Qy
k
.
We select for the algorithm the most favorable step size, that is, the one that takes us from y
k
to the lowest point in
the direction of p
k
. This can be found by differentiating the function
e(y
k
+αp
k
) =
1
2
(y
k
+αp
k
)
T
Q(y
k
+αp
k
)
with respect to α, and setting the derivative to zero to obtain the optimal step α
k
. We have
∂e(y
k
+αp
k
)
∂α
= (y
k
+αp
k
)
T
Qp
k
and setting this to zero yields
α
k
= −
(Qy
k
)
T
p
k
p
T
k
Qp
k
= −
g
T
k
p
k
p
T
k
Qp
k
= g
k

p
T
k
p
k
p
T
k
Qp
k
= g
k

g
T
k
g
k
g
T
k
Qg
k
. (4.4)
42 CHAPTER 4. FUNCTION OPTIMIZATION
Thus, the basic step of our steepest descent can be written as follows:
y
k+1
= y
k
+g
k

g
T
k
g
k
g
T
k
Qg
k
p
k
that is,
y
k+1
= y
k

g
T
k
g
k
g
T
k
Qg
k
g
k
. (4.5)
How much closer did this step bring us to the solution y

= 0? In other words, how much smaller is e(y
k+1
),
relative to the value e(y
k
) at the previous step? The answer is, often not much, as we shall now prove. The arguments
and proofs below are adapted from D. G. Luenberger, Introduction to Linear and Nonlinear Programming, Addison-
Wesley, 1973.
From the definition of e and from equation (4.5) we obtain
e(y
k
) −e(y
k+1
)
e(y
k
)
=
y
T
k
Qy
k
−y
T
k+1
Qy
k+1
y
T
k
Qy
k
=
y
T
k
Qy
k

_
y
k

g
T
k
g
k
g
T
k
Qg
k
g
k
_
T
Q
_
y
k

g
T
k
g
k
g
T
k
Qg
k
g
k
_
y
T
k
Qy
k
=
2
g
T
k
g
k
g
T
k
Qg
k
g
T
k
Qy
k

_
g
T
k
g
k
g
T
k
Qg
k
_
2
g
T
k
Qg
k
y
T
k
Qy
k
=
2g
T
k
g
k
g
T
k
Qy
k
−(g
T
k
g
k
)
2
y
T
k
Qy
k
g
T
k
Qg
k
.
Since Q is invertible we have
g
k
= Qy
k
⇒ y
k
= Q
−1
g
k
and
y
T
k
Qy
k
= g
T
k
Q
−1
g
k
so that
e(y
k
) −e(y
k+1
)
e(y
k
)
=
(g
T
k
g
k
)
2
g
T
k
Q
−1
g
k
g
T
k
Qg
k
.
This can be rewritten as follows by rearranging terms:
e(y
k+1
) =
_
1 −
(g
T
k
g
k
)
2
g
T
k
Q
−1
g
k
g
T
k
Qg
k
_
e(y
k
) (4.6)
so if we can bound the expression in parentheses we have a bound on the rate of convergence of steepest descent. To
this end, we introduce the following result.
Lemma 4.1.1 (Kantorovich inequality) Let Qbe a positive definite, symmetric, n×n matrix. For any vector y there
holds
(y
T
y)
2
y
T
Q
−1
y y
T
Qy


1
σ
n

1

n
)
2
where σ
1
and σ
n
are, respectively, the largest and smallest singular values of Q.
4.1. LOCAL MINIMIZATION AND STEEPEST DESCENT 43
Proof. Let
Q = UΣU
T
be the singular value decomposition of the symmetric (hence V = U) matrix Q. Because Q is positive definite, all its
singular values are strictly positive, since the smallest of them satisfies
σ
n
= min
y=1
y
T
Qy > 0
by the definition of positive definiteness. If we let
z = U
T
y
we have
(y
T
y)
2
y
T
Q
−1
y y
T
Qy
=
(y
T
U
T
Uy)
2
y
T

−1
U
T
y y
T
UΣU
T
y
=
(z
T
z)
2
z
T
Σ
−1
z z
T
Σz
=
1/

n
i=1
θ
i
σ
i

n
i=1
θ
i

i
=
φ(σ)
ψ(σ)
(4.7)
where the coefficients
θ
i
=
z
2
i
z
2
add up to one. If we let
σ =
n

i=1
θ
i
σ
i
, (4.8)
then the numerator φ(σ) in (4.7) is 1/σ. Of course, there are many ways to choose the coefficients θ
i
to obtain a
particular value of σ. However, each of the singular values σ
j
can be obtained by letting θ
j
= 1 and all other θ
i
to
zero. Thus, the values 1/σ
j
for j = 1, . . . , n are all on the curve 1/σ. The denominator ψ(σ) in (4.7) is a convex
combination of points on this curve. Since 1/σ is a convex function of σ, the values of the denominator ψ(σ) of (4.7)
must be in the shaded area in figure 4.1. This area is delimited from above by the straight line that connects point

1
, 1/σ
1
) with point (σ
n
, 1/σ
n
), that is, by the line with ordinate
λ(σ) = (σ
1

n
−σ)/(σ
1
σ
n
) .
For the same vector of coefficients θ
i
, the values of φ(σ), ψ(σ), and λ(σ) are on the vertical line corresponding to
the value of σ given by (4.8). Thus an appropriate bound is
φ(σ)
ψ(σ)
≥ min
σ
1
≤σ≤σ
n
φ(σ)
λ(σ)
= min
σ
1
≤σ≤σ
n
1/σ

1

n
−σ)/(σ
1
σ
n
)
.
The minimum is achieved at σ = (σ
1

n
)/2, yielding the desired result. ∆
Thanks to this lemma, we can state the main result on the convergence of the method of steepest descent.
Theorem 4.1.2 Let
f(x) = c + a
T
x +
1
2
x
T
Qx
be a quadratic function of x, with Q symmetric and positive definite. For any x
0
, the method of steepest descent
x
k+1
= x
k

g
T
k
g
k
g
T
k
Qg
k
g
k
(4.9)
where
g
k
= g(x
k
) =
∂f
∂x
¸
¸
¸
¸
x=x
k
= a +Qx
k
44 CHAPTER 4. FUNCTION OPTIMIZATION
σ
2 1
σ σ σ
n
λ(σ)
φ(σ)
ψ(σ)
φ,ψ,λ
σ
Figure 4.1: Kantorovich inequality.
converges to the unique minimum point
x

= −Q
−1
a
of f. Furthermore, at every step k there holds
f(x
k+1
) −f(x

) ≤
_
σ
1
−σ
n
σ
1

n
_
2
(f(x
k
) −f(x

))
where σ
1
and σ
n
are, respectively, the largest and smallest singular value of Q.
Proof. From the definitions
y = x −x

and e(y) =
1
2
y
T
Qy (4.10)
we immediately obtain the expression for steepest descent in terms of f and x. By equations (4.3) and (4.6) and the
Kantorovich inequality we obtain
f(x
k+1
) −f(x

) = e(y
k+1
) =
_
1 −
(g
T
k
g
k
)
2
g
T
k
Q
−1
g
k
g
T
k
Qg
k
_
e(y
k
) ≤
_
1 −

1
σ
n

1

n
)
2
_
e(y
k
) (4.11)
=
_
σ
1
−σ
n
σ
1

n
_
2
(f(x
k
) −f(x

)) . (4.12)
Since the ratio in the last term is smaller than one, it follows immediately that f(x
k
) − f(x

) → 0 and hence, since
the minimum of f is unique, that x
k
→x

. ∆
The ratio κ(Q) = σ
1

n
is called the condition number of Q. The larger the condition number, the closer the
fraction (σ
1
− σ
n
)/(σ
1
+ σ
n
) is to unity, and the slower convergence. It is easily seen why this happens in the case
4.1. LOCAL MINIMIZATION AND STEEPEST DESCENT 45
p
0
x
*
x
0 1
p
Figure 4.2: Trajectory of steepest descent.
in which x is a two-dimensional vector, as in figure 4.2, which shows the trajectory x
k
superimposed on a set of
isocontours of f(x).
There is one good, but very precarious case, namely, when the starting point x
0
is at one apex (tip of either axis)
of an isocontour ellipse. In that case, one iteration will lead to the minimum x

. In all other cases, the line in the
direction p
k
of steepest descent, which is orthogonal to the isocontour at x
k
, will not pass through x

. The minimum
of f along that line is tangent to some other, lower isocontour. The next step is orthogonal to the latter isocontour (that
is, parallel to the gradient). Thus, at every step the steepest descent trajectory is forced to make a ninety-degree turn.
If isocontours were circles (σ
1
= σ
n
) centered at x

, then the first turn would make the new direction point to x

, and
minimization would get there in just one more step. This case, in which κ(Q) = 1, is consistent with our analysis,
because then
σ
1
−σ
n
σ
1

n
= 0 .
The more elongated the isocontours, that is, the greater the condition number κ(Q), the farther away a line orthogonal
to an isocontour passes from x

, and the more steps are required for convergence.
For general (that is, non-quadratic) f, the analysis above applies once x
k
gets close enough to the minimum, so
that f is well approximated by a paraboloid. In this case, Q is the matrix of second derivatives of f with respect to x,
and is called the Hessian of f. In summary, steepest descent is good for functions that have a well conditioned Hessian
near the minimum, but can become arbitrarily slow for poorly conditioned Hessians.
To characterize the speed of convergence of different minimization algorithms, we introduce the notion of the
order of convergence. This is defined as the largest value of q for which the
lim
k→∞
x
k+1
−x

x
k
−x

q
is finite. If β is this limit, then close to the solution (that is, for large values of k) we have
x
k+1
−x

≈ βx
k
−x

q
for a minimization method of order q. In other words, the distance of x
k
from x

is reduced by the q-th power at every
step, so the higher the order of convergence, the better. Theorem 4.1.2 implies that steepest descent has at best a linear
order of convergence. In fact, the residuals |f(x
k
) − f(x

)| in the values of the function being minimized converge
46 CHAPTER 4. FUNCTION OPTIMIZATION
linearly. Since the gradient of f approaches zero when x
k
tends to x

, the arguments x
k
to f can converge to x

even
more slowly.
To complete the steepest descent algorithm we need to specify how to check whether a minimum has been reached.
One criterion is to check whether the value of f(x
k
) has significantly decreased from f(x
k−1
). Another is to check
whether x
k
is significantly different from x
k−1
. Close to the minimum, the derivatives of f are close to zero, so
|f(x
k
) − f(x
k−1
)| may be very small but x
k
− x
k−1
may still be relatively large. Thus, the check on x
k
is more
stringent, and therefore preferable in most cases. In fact, usually one is interested in the value of x

, rather than in that
of f(x

). In summary, the steepest descent algorithm can be stopped when
x
k
−x
k−1
<
where the positive constant is provided by the user.
In our analysis of steepest descent, we used the Hessian Qin order to compute the optimal step size α (see equation
(4.4)). We used Qbecause it was available, but its computation during steepest descent would in general be overkill. In
fact, only gradient information is necessary to find p
k
, and a line search in the direction of p
k
can be used to determine
the step size α
k
. In contrast, the Hessian of f(x) requires computing
_
n
2
_
second derivatives if x is an n-dimensional
vector.
Using line search to find α
k
guarantees that a minimum in the direction p
k
is actually reached even when the
parabolic approximation is inadequate. Here is how line search works.
Let
h(α) = f(x
k
+αp
k
) (4.13)
be the scalar function of one variable that is obtained by restricting the function f to the line through the current point
x
k
and in the direction of p
k
. Line search first determines two points a, c that bracket the desired minimum α
k
, in the
sense that a ≤ α
k
≤ c, and then picks a point between a and c, say, b = (a + c)/2. The only difficulty here is to
find c. In fact, we can set a = 0, corresponding through equation (4.13) to the starting point x
k
. A point c that is on
the opposite side of the minimum with respect to a can be found by increasing α through values α
1
= a, α
2
, . . . until
h(α
i
) is greater than h(α
i−1
). Then, if we can assume that h is convex between α
1
and α
i
, we can set c = α
i
. In
fact, the derivative of h at a is negative, so the function is initially decreasing, but it is increasing between α
i−1
and
α
i
= c, so the minimum must be somewhere between a and c. Of course, if we cannot assume convexity, we may find
the wrong minimum, but there is no general-purpose fix to this problem.
Line search now proceeds by shrinking the bracketing triple (a, b, c) until c−a is smaller than the desired accuracy
in determining α
k
. Shrinking works as follows:
if b −a > c −b
u = (a +b)/2
if f(u) > f(b)
(a, b, c) = (u, b, c)
otherwise
(a, b, c) = (a, u, b)
end
otherwise
u = (b +c)/2
if f(u) > f(b)
(a, b, c) = (a, b, u)
otherwise
(a, b, c) = (b, u, c)
end
end.
4.2. NEWTON’S METHOD 47
It is easy to see that in each case the bracketing triple (a, b, c) preserves the property that f(b) ≤ f(a) and
f(b) ≤ f(c), and therefore the minimum is somewhere between a and c. In addition, at every step the interval (a, c)
shrinks to 3/4 of its previous size, so line search will find the minimum in a number of steps that is logarithmic in the
desired accuracy.
4.2 Newton’s Method
If a function can be well approximated by a paraboloid in the region in which minimization is performed, the analysis
in the previous section suggests a straight-forward fix to the slow convergence of steepest descent. In fact, equation
(4.2) tells us how to jump in one step from the starting point x
0
to the minimum x

. Of course, when f(x) is not
exactly a paraboloid, the new value x
1
will be different from x

. Consequently, iterations are needed, but convergence
can be expected to be faster. This is the idea of Newton’s method, which we now summarize. Let
f(x
k
+ ∆x) ≈ f(x
k
) + g
T
k
∆x +
1
2
∆x
T
Q
k
∆x (4.14)
be the first terms of the Taylor series expansion of f about the current point x
k
, where
g
k
= g(x
k
) =
∂f
∂x
¸
¸
¸
¸
x=x
k
and
Q
k
= Q(x
k
) =

2
f
∂x∂x
T
¸
¸
¸
¸
x=x
k
=
_
¸
¸
_

2
f
∂x
2
1
· · ·

2
f
∂x1∂xn
.
.
.
.
.
.

2
f
∂xn∂x1
· · ·

2
f
∂x
2
n
_
¸
¸
_
x=x
k
are the gradient and Hessian of f evaluated at the current point x
k
. Notice that even when f is a paraboloid, the
gradient g
k
is different from a as used in equation (4.1). In fact, a and Q are the coefficients of the Taylor expansion
of f around point x = 0, while g
k
and Q
k
are the coefficients of the Taylor expansion of f around the current point
x
k
. In other words, gradient and Hessian are constantly reevaluated in Newton’s method.
To the extent that approximation (4.14) is valid, we can set the derivatives of f(x
k
+ ∆x) with respect to ∆x to
zero, and obtain, analogously to equation (4.2), the linear system
Q
k
∆x = −g
k
, (4.15)
whose solution ∆x
k
= α
k
p
k
yields at the same time the step direction p
k
= ∆x
k
/∆x
k
and the step size α
k
=
∆x
k
. The direction is of course undefined once the algorithm has reached a minimum, that is, when α
k
= 0.
A minimization algorithm in which the step direction p
k
and size α
k
are defined in this manner is called Newton’s
method. The corresponding p
k
is termed the Newton direction, and the step defined by equation (4.15) is the Newton
step.
The greater speed of Newton’s method over steepest descent is borne out by analysis: while steepest descent has a
linear order of convergence, Newton’s method is quadratic. In fact, let
y(x) = x −Q(x)
−1
g(x)
be the place reached by a Newton step starting at x (see equation (4.15)), and suppose that at the minimum x

the
Hessian Q(x

) is nonsingular. Then
y(x

) = x

because g(x

) = 0, and
x
k+1
−x

= y(x
k
) −x

= y(x
k
) −y(x

) .
48 CHAPTER 4. FUNCTION OPTIMIZATION
From the mean-value theorem, we have
x
k+1
−x

= y(x
k
) −y(x

) ≤
_
_
_
_
_
∂y
∂x
T
_
x=x

(x
k
−x

)
_
_
_
_
+
1
2
¸
¸
¸
¸

2
y
∂x∂x
T
¸
¸
¸
¸
x=
ˆ
x
x
k
−x

2
where ˆx is some point on the line between x

and x
k
. Since y(x

) = x

, the first derivatives of y at x

are zero, so that
the first term in the right-hand side above vanishes, and
x
k+1
−x

≤ c x
k
−x

2
where c depends on third-order derivatives of f near x

. Thus, the convergence rate of Newton’s method is of order at
least two.
For a quadratic function, as in equation (4.1), steepest descent takes many steps to converge, while Newton’s
method reaches the minimum in one step. However, this single iteration in Newton’s method is more expensive,
because it requires both the gradient g
k
and the Hessian Q
k
to be evaluated, for a total of n +
_
n
2
_
derivatives. In
addition, the Hessian must be inverted, or, at least, system (4.15) must be solved. For very large problems, in which
the dimension n of x is thousands or more, storing and manipulating a Hessian can be prohibitive. In contrast, steepest
descent requires the gradient g
k
for selecting the step direction p
k
, and a line search in the direction p
k
to find the
step size. The method of conjugate gradients, discussed in the next section, is motivated by the desire to accelerate
convergence with respect to the steepest descent method, but without paying the storage cost of Newton’s method.
4.3 Conjugate Gradients
Newton’s method converges faster (quadratically) than steepest descent (linear convergence rate) because it uses more
information about the function f being minimized. Steepest descent locally approximates the function with planes,
because it only uses gradient information. All it can do is to go downhill. Newton’s method approximates f with
paraboloids, and then jumps at every iteration to the lowest point of the current approximation. The bottom line is that
fast convergence requires work that is equivalent to evaluating the Hessian of f.
Prima facie, the method of conjugate gradients discussed in this section seems to violate this principle: it achieves
fast, superlinear convergence, similarly to Newton’s method, but it only requires gradient information. This paradox,
however, is only apparent. Conjugate gradients works by taking n steps for each of the steps in Newton’s method.
It effectively solves the linear system (4.2) of Newton’s method, but it does so by a sequence of n one-dimensional
minimizations, each requiring one gradient computation and one line search.
Overall, the work done by conjugate gradients is equivalent to that done by Newton’s method. However, system
(4.2) is never constructed explicitly, and the matrix Q is never stored. This is very important in cases where x has
thousands or even millions of components. These high-dimensional problems arise typically from the discretization
of partial differential equations. Say for instance that we want to compute the motion of points in an image as a
consequence of camera motion. Partial differential equations relate image intensities over space and time to the motion
of the underlying image features. At every pixel in the image, this motion, called the motion field, is represented by
a vector whose magnitude and direction describe the velocity of the image feature at that pixel. Thus, if an image
has, say, a quarter of a million pixels, there are n = 500, 000 unknown motion field values. Storing and inverting a
500, 000 ×500, 000 Hessian is out of the question. In cases like these, conjugate gradients saves the day.
The conjugate gradients method described in these notes is the so-called Polak-Ribi` ere variation. It will be intro-
duced in three steps. First, it will be developed for the simple case of minimizing a quadratic function with positive-
definite and known Hessian. This quadratic function f(x) was introduced in equation (4.1). We know that in this case
minimizing f(x) is equivalent to solving the linear system (4.2). Rather than an iterative method, conjugate gradients
is a direct method for the quadratic case. This means that the number of iterations is fixed. Specifically, the method
converges to the solution in n steps, where n is the number of components of x. Because of the equivalence with
a linear system, conjugate gradients for the quadratic case can also be seen as an alternative method for solving a
linear system, although the version presented here will only work if the matrix of the system is symmetric and positive
definite.
4.3. CONJUGATE GRADIENTS 49
Second, the assumption that the Hessian Q in expression (4.1) is known will be removed. As discussed above, this
is the main reason for using conjugate gradients.
Third, the conjugate gradients method will be extended to general functions f(x). In this case, the method is no
longer direct, but iterative, and the cost of finding the minimum depends on the desired accuracy. This occurs because
the Hessian of f is no longer a constant, as it was in the quadratic case. As a consequence, a certain property that holds
in the quadratic case is now valid only approximately. In spite of this, the convergence rate of conjugate gradients is
superlinear, somewhere between Newton’s method and steepest descent. Finding tight bounds for the convergence rate
of conjugate gradients is hard, and we will omit this proof. We rely instead on the intuition that conjugate gradients
solves system (4.2), and that the quadratic approximation becomes more and more valid as the algorithm converges
to the minimum. If the function f starts to behave like a quadratic function early, that is, if f is nearly quadratic in a
large neighborhood of the minimum, convergence is fast, as it requires close to the n steps that are necessary in the
quadratic case, and each of the steps is simple. This combination of fast convergence, modest storage requirements,
and low computational cost per iteration explains the popularity of conjugate gradients methods for the optimization
of functions of a large number of variables.
4.3.1 The Quadratic Case
Suppose that we want to minimize the quadratic function
f(x) = c + a
T
x +
1
2
x
T
Qx (4.16)
where Q is a symmetric, positive definite matrix, and x has n components. As we saw in our discussion of steepest
descent, the minimum x

is the solution to the linear system
Qx = −a . (4.17)
We know how to solve such a system. However, all the methods we have seen so far involve explicit manipulation
of the matrix Q. We now consider an alternative solution method that does not need Q, but only the quantity
g
k
= Qx
k
+ a
that is, the gradient of f(x), evaluated at n different points x
1
, . . . , x
n
. We will see that the conjugate gradients method
requires n gradient evaluations and n line searches in lieu of each n ×n matrix inversion in Newton’s method.
Formal proofs can be found in Elijah Polak, Optimization — Algorithms and consistent approximations, Springer,
NY, 1997. The arguments offered below appeal to intuition.
Consider the case n = 3, in which the variable x in f(x) is a three-dimensional vector. Then the quadratic function
f(x) is constant over ellipsoids, called isosurfaces, centered at the minimum x

. How can we start from a point x
0
on one of these ellipsoids and reach x

by a finite sequence of one-dimensional searches? In connection with steepest
descent, we noticed that for poorly conditioned Hessians orthogonal directions lead to many small steps, that is, to
slow convergence.
When the ellipsoids are spheres, on the other hand, this works much better. The first step takes from x
0
to x
1
, and
the line between x
0
and x
1
is tangent to an isosurface at x
1
. The next step is in the direction of the gradient, so that
the new direction p
1
is orthogonal to the previous direction p
0
. This would then take us to x

right away. Suppose
however that we cannot afford to compute this special direction p
1
orthogonal to p
0
, but that we can only compute
some direction p
1
orthogonal to p
0
(there is an n − 1-dimensional space of such directions!). It is easy to see that in
that case n steps will take us to x

. In fact, since isosurfaces are spheres, each line minimization is independent of the
others: The first step yields the minimum in the space spanned by p
0
, the second step then yields the minimum in the
space spanned by p
0
and p
1
, and so forth. After n steps we must be done, since p
0
. . . , p
n−1
span the whole space.
In summary, any set of orthogonal directions, with a line search in each direction, will lead to the minimum for
spherical isosurfaces. Given an arbitrary set of ellipsoidal isosurfaces, there is a one-to-one mapping with a spherical
system: if Q = UΣU
T
is the SVD of the symmetric, positive definite matrix Q, then we can write
1
2
x
T
Qx =
1
2
y
T
y
50 CHAPTER 4. FUNCTION OPTIMIZATION
where
y = Σ
1/2
U
T
x . (4.18)
Consequently, there must be a condition for the original problem (in terms of Q) that is equivalent to orthogonality for
the spherical problem. If two directions q
i
and q
j
are orthogonal in the spherical context, that is, if
q
T
i
q
j
= 0 ,
what does this translate into in terms of the directions p
i
and p
j
for the ellipsoidal problem? We have
q
i,j
= Σ
1/2
U
T
p
i,j
,
so that orthogonality for q
i,j
becomes
p
T
i

1/2
Σ
1/2
U
T
p
j
= 0
or
p
T
i
Qp
j
= 0 . (4.19)
This condition is called Q-conjugacy, or Q-orthogonality: if equation (4.19) holds, then p
i
and p
j
are said to be
Q-conjugate or Q-orthogonal to each other. We will henceforth simply say “conjugate” for brevity.
In summary, if we can find n directions p
0
, . . . , p
n−1
that are mutually conjugate, and if we do line minimization
along each direction p
k
, we reach the minimum in at most n steps. Of course, we cannot use the transformation (4.18)
in the algorithm, because Σ and especially U
T
are too large. So now we need to find a method for generating n
conjugate directions without using either Q or its SVD. We do this in two steps. First, we find conjugate directions
whose definitions do involve Q. Then, in the next subsection, we rewrite these expressions without Q.
Here is the procedure, due to Hestenes and Stiefel (Methods of conjugate gradients for solving linear systems, J.
Res. Bureau National Standards, section B, Vol 49, pp. 409-436, 1952), which also incorporates the steps from x
0
to
x
n
:
g
0
= g(x
0
)
p
0
= −g
0
for k = 0 . . . , n −1
α
k
= arg min
α≥0
f(x
k
+αp
k
)
x
k+1
= x
k

k
p
k
g
k+1
= g(x
k+1
)
γ
k
=
g
T
k+1
Qp
k
p
T
k
Qp
k
p
k+1
= −g
k+1

k
p
k
end
where
g
k
= g(x
k
) =
∂f
∂x
¸
¸
¸
¸
x=x
k
is the gradient of f at x
k
.
It is simple to see that p
k
and p
k+1
are conjugate. In fact,
p
T
k
Qp
k+1
= p
T
k
Q(−g
k+1

k
p
k
)
= −p
T
k
Qg
k+1
+
g
T
k+1
Qp
k
p
T
k
Qp
k
p
T
k
Qp
k
= −p
T
k
Qg
k+1
+ g
T
k+1
Qp
k
= 0 .
It is somewhat more cumbersome to show that p
i
and p
k+1
for i = 0, . . . , k are also conjugate. This can be done by
induction. The proof is based on the observation that the vectors p
k
are found by a generalization of Gram-Schmidt
(theorem 2.4.2) to produce conjugate rather than orthogonal vectors. Details can be found in Polak’s book mentioned
earlier.
4.3. CONJUGATE GRADIENTS 51
4.3.2 Removing the Hessian
The algorithm shown in the previous subsection is a correct conjugate gradients algorithm. However, it is computa-
tionally inadequate because the expression for γ
k
contains the Hessian Q, which is too large. We now show that γ
k
can be rewritten in terms of the gradient values g
k
and g
k+1
only. To this end, we notice that
g
k+1
= g
k

k
Qp
k
,
or
α
k
Qp
k
= g
k+1
−g
k
.
In fact,
g(x) = a +Qx
so that
g
k+1
= g(x
k+1
) = g(x
k

k
p
k
) = a +Q(x
k

k
p
k
) = g
k

k
Qp
k
.
We can therefore write
γ
k
=
g
T
k+1
Qp
k
p
T
k
Qp
k
=
g
T
k+1
α
k
Qp
k
p
T
k
α
k
Qp
k
=
g
T
k+1
(g
k+1
−g
k
)
p
T
k
(g
k+1
−g
k
)
,
and Q has disappeared.
This expression for γ
k
can be further simplified by noticing that
p
T
k
g
k+1
= 0
because the line along p
k
is tangent to an isosurface at x
k+1
, while the gradient g
k+1
is orthogonal to the isosurface at
x
k+1
. Similarly,
p
T
k−1
g
k
= 0 .
Then, the denominator of γ
k
becomes
p
T
k
(g
k+1
−g
k
) = −p
T
k
g
k
= (g
k
−γ
k−1
p
k−1
)
T
g
k
= g
T
k
g
k
.
In conclusion, we obtain the Polak-Ribi` ere formula
γ
k
=
g
T
k+1
(g
k+1
−g
k
)
g
T
k
g
k
.
4.3.3 Extension to General Functions
We now know how to minimize the quadratic function (4.16) in n steps, without ever constructing the Hessian explic-
itly. When the function f(x) is arbitrary, the same algorithm can be used.
However, n iterations will not suffice. In fact, the Hessian, which was constant for the quadratic case, now is a
function of x
k
. Strictly speaking, we then lose conjugacy, since p
k
and p
k+1
are associated to different Hessians.
However, as the algorithm approaches the minimum x

, the quadratic approximation becomes more and more valid,
and a few cycles of n iterations each will achieve convergence.
52 CHAPTER 4. FUNCTION OPTIMIZATION
Chapter 5
Eigenvalues and Eigenvectors
Given a linear transformation
b = Ax ,
the singular value decomposition A = UΣV
T
of A transforms the domain of the transformation via the matrix V
T
and its range via the matrix U
T
so that the transformed system is diagonal. In fact, the equation b = UΣV
T
x can be
written as follows
U
T
b = ΣV
T
x ,
that is,
c = Σy
where
y = V
T
x and c = U
T
b ,
and where Σ is diagonal. This is a fundamental transformation to use whenever the domain and the range of A are
separate spaces. Often, however, domain and range are intimately related to one another even independently of the
transformation A. The most important example is perhaps that of a system of linear differential equations, of the form
˙ x = Ax
where Ais n×n. For this equation, the fact that Ais square is not a coincidence. In fact, x is assumed to be a function
of some real scalar variable t (often time), and ˙ x is the derivative of x with respect to t:
˙ x =
dx
dt
.
In other words, there is an intimate, pre-existing relation between x and ˙ x, and one cannot change coordinates for x
without also changing those for ˙ x accordingly. In fact, if V is an orthogonal matrix and we define
y = V
T
x ,
then the definition of ˙ x forces us to transform ˙ x by V
T
as well:
d V
T
x
dt
= V
T
dx
dt
= V
T
˙ x .
In brief, the SVD does nothing useful for systems of linear differential equations, because it diagonalizes Aby two
different transformations, one for the domain and one for the range, while we need a single transformation. Ideally,
we would like to find an orthogonal matrix S and a diagonal matrix Λ such that
A = SΛS
T
(5.1)
53
54 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
so that if we define
y = S
T
x
we can write the equivalent but diagonal differential system
˙ y = Λy .
This is now much easier to handle, because it is a system of n independent, scalar differential equations, which can be
solved separately. The solutions can then be recombined through
x = Sy .
We will see all of this in greater detail soon.
Unfortunately, writing A in the form (5.1) is not always possible. This stands to reason, because now we are
imposing stronger constraints on the terms of the decomposition. It is like doing an SVD but with the additional
constraint U = V . If we refer back to figure 3.1, now the circle and the ellipse live in the same space, and the
constraint U = V implies that the vectors v
i
on the circle that map into the axes σ
i
u
i
of the ellipse are parallel to the
axes themselves. This will only occur for very special matrices.
In order to make a decomposition like (5.1) possible, we weaken the constraints in several ways:
• the elements of S and Λ are allowed to be complex, rather than real;
• the elements on the diagonal of Λ are allowed to be negative; in fact, they can be even non-real;
• S is required to be only invertible, rather than orthogonal.
To distinguish invertible from orthogonal matrices we use the symbol Q for invertible and S for orthogonal. In some
cases, it will be possible to diagonalize A by orthogonal transformations S and S
T
. Finally, for complex matrices we
generalize the notion of transpose by introducing the Hermitian operator: The matrix Q
H
(pronounced “QHermitian”)
is defined to be the complex conjugate of the transpose of Q. If Q happens to be real, conjugate transposition becomes
simply transposition, so the Hermitian is a generalization of the transpose. A matrix S is said to be unitary if
S
H
S = SS
H
= I ,
so unitary generalizes orthogonal for complex matrices. Unitary matrices merely rotate or flip vectors, in the sense
that they do not alter the vectors’ norms. For complex vectors, the norm squared is defined as
x
2
= x
H
x ,
and if S is unitary we have
Sx
2
= x
H
S
H
Sx = x
H
x = x
2
.
Furthermore, if x
1
and x
2
are mutually orthogonal, in the sense that
x
H
1
x
2
= 0 ,
then Sx
1
and Sx
2
are orthogonal as well:
x
H
1
S
H
Sx
2
= x
H
1
x
2
= 0 .
In contrast, a nonunitary transformation Q can change the norms of vectors, as well as the inner products between
vectors. A matrix that is equal to its Hermitian is called a Hermitian matrix.
In summary, in order to diagonalize a square matrix A from a system of linear differential equations we generally
look for a decomposition of A of the form
A = QΛQ
−1
(5.2)
55
where Q and Λ are complex, Q is invertible, and Λ is diagonal. For some special matrices, this may specialize to
A = SΛS
H
with unitary S.
Whenever two matrices A and B, diagonal or not, are related by
A = QBQ
−1
,
they are said to be similar to each other, and the transformation of B into A (and vice versa) is called a similarity
transformation.
The equation A = QΛQ
−1
can be rewritten as follows:
AQ = QΛ
or separately for every column of Q as follows:
Aq
i
= λ
i
q
i
(5.3)
where
Q =
_
q
1
· · · q
n
¸
and Λ = diag(λ
1
, . . . , λ
n
) .
Thus, the columns of q
i
of Q and the diagonal entries λ
i
of Λ are solutions of the eigenvalue/eigenvector equation
Ax = λx , (5.4)
which is how eigenvalues and eigenvectors are usually introduced. In contrast, we have derived this equation from the
requirement of diagonalizing a matrix by a similarity transformation. The columns of Q are called eigenvectors, and
the diagonal entries of Λ are called eigenvalues.
−2 −1 0 1 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
Figure 5.1: Effect of the transformation (5.5) on a sample of points on the unit circle. The dashed lines are vectors that
do not change direction under the transformation.
That real eigenvectors and eigenvalues do not always exist can be clarified by considering the eigenvalue problem
from a geometrical point of view in the n = 2 case. As we know, an invertible linear transformation transforms the
unit circle into an ellipse. Each point on the unit circle is transformed into some point on the ellipse. Figure 5.1 shows
the effect of the transformation represented by the matrix
A =
_
2/3 4/

3
0 2
_
(5.5)
56 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
for a sample of points on the unit circle. Notice that there are many transformations that map the unit circle into the
same ellipse. In fact, the circle in figure 5.1 can be rotated, pulling the solid lines along. Each rotation yields another
matrix A, but the resulting ellipse is unchanged. In other words, the curve-to-curve transformation from circle to
ellipse is unique, but the point-to-point transformation is not. Matrices represent point-to-point transformations.
The eigenvalue problem amounts to finding axes q
1
, q
2
that are mapped into themselves by the original transfor-
mation A (see equation (5.3)). In figure 5.1, the two eigenvectors are shown as dashed lines. Notice that they do not
correspond to the axes of the ellipse, and that they are not orthogonal. Equation (5.4) is homogeneous in x, so x can
be assumed to be a unit vector without loss of generality.
Given that the directions of the input vectors are generally changed by the transformation A, as evident from figure
5.1, it is not obvious whether the eigenvalue problem admits a solution at all. We will see that the answer depends
on the matrix A, and that a rather diverse array of situations may arise. In some cases, the eigenvalues and their
eigenvectors exist, but they are complex. The geometric intuition is hidden, and the problem is best treated as an
algebraic one. In other cases, all eigenvalues exist, perhaps all real, but not enough eigenvectors can be found, and the
matrix A cannot be diagonalized. In particularly good cases, there are n real, orthonormal eigenvectors. In bad cases,
we have to give up the idea of diagonalizing A, and we can only triangularize it. This turns out to be good enough for
solving linear differential systems, just as triangularization was sufficient for solving linear algebraic systems.
5.1 Computing Eigenvalues and Eigenvectors Algebraically
Let us rewrite the eigenvalue equation
Ax = λx
as follows:
(A−λI)x = 0 . (5.6)
This is a homogeneous, square system of equations, which admits nontrivial solutions iff the matrix A − λI is rank-
deficient. A square matrix B is rank-deficient iff its determinant,
det(B) =
_
b
11
if B is 1 ×1

n
i=1
(−1)
i+1
b
i1
det(B
i1
) otherwise
is zero. In this expression, B
ij
is the algebraic complement of entry b
ij
, defined as the (n − 1) × (n − 1) matrix
obtained by removing row i and column j from B.
Volumes have been written about the properties of the determinant. For our purposes, it is sufficient to recall the
following properties from linear algebra:
• det(B) = det(B
T
);
• det(
_
b
1
· · · b
n
¸
) = 0 iff b
1
, . . . , b
n
are linearly dependent;
• det(
_
b
1
· · · b
i
· · · b
j
· · · b
n
¸
) = −det(
_
b
1
· · · b
j
· · · b
i
· · · b
n
¸
);
• det(BC) = det(B) det(C).
Thus, for system (5.6) to admit nontrivial solutions, we need
det(A−λI) = 0 . (5.7)
From the definition of determinant, it follows, by very simple induction, that the left-hand side of equation (5.7)
is a polynomial of degree n in λ, and that the coefficient of λ
n
is 1. Therefore, equation (5.7), which is called the
characteristic equation of A, has n complex solutions, in the sense that
det(A−λI) = (−1)
n
(λ −λ
1
) · . . . · (λ −λ
n
)
where some of the λ
i
may coincide. In other words, an n × n matrix has at most n distinct eigenvalues. The case of
exactly n distinct eigenvalues is of particular interest, because of the following results.
5.1. COMPUTING EIGENVALUES AND EIGENVECTORS ALGEBRAICALLY 57
Theorem 5.1.1 Eigenvectors x
1
, . . . , x
k
corresponding to distinct eigenvalues λ
1
, . . . , λ
k
are linearly independent.
Proof. Suppose that c
1
x
1
+. . . +c
k
x
k
= 0 where the x
i
are eigenvectors of a matrix A. We need to show that then
c
1
= . . . = c
k
= 0. By multiplying by A we obtain
c
1
Ax
1
+. . . +c
k
Ax
k
= 0
and because x
1
, . . . , x
k
are eigenvectors corresponding to eigenvalues λ
1
, . . . , λ
k
, we have
c
1
λ
1
x
1
+. . . +c
k
λ
k
x
k
= 0 . (5.8)
However, from
c
1
x
1
+. . . +c
k
x
k
= 0
we also have
c
1
λ
k
x
1
+. . . +c
k
λ
k
x
k
= 0
and subtracting this equation from equation (5.8) we have
c
1

1
−λ
k
)x
1
+. . . +c
k−1

k−1
−λ
k
)x
k−1
= 0 .
Thus, we have reduced the summation to one containing k − 1 terms. Since all λ
i
are distinct, the differences in
parentheses are all nonzero, and we can replace each x
i
by x

i
= (λ
i
−λ
k
)x
i
, which is still an eigenvector of A:
c
1
x

1
+. . . +c
k−1
x

k−1
= 0 .
We can repeat this procedure until only one term remains, and this forces c
1
= 0, so that
c
2
x
2
+. . . +c
k
x
k
= 0
This entire argument can be repeated for the last equation, therefore forcing c
2
= 0, and so forth.
In summary, the equation c
1
x
1
+. . . +c
k
x
k
= 0 implies that c
1
= . . . = c
k
= 0, that is, that the vectors x
1
, . . . , x
k
are linearly independent. ∆
For Hermitian matrices (and therefore for real symmetric matrices as well), the situation is even better.
Theorem 5.1.2 A Hermitian matrix has real eigenvalues.
Proof. A matrix A is Hermitian iff A = A
H
. Let λ and x be an eigenvalue of A and a corresponding eigenvector:
Ax = λx . (5.9)
By taking the Hermitian we obtain
x
H
A
H
= λ

x
H
.
Since A = A
H
, the last equation can be rewritten as follows:
x
H
A = λ

x
H
. (5.10)
If we multiply equation (5.9) from the left by x
H
and equation (5.10) from the right by x, we obtain
x
H
Ax = λx
H
x
x
H
Ax = λ

x
H
x
which implies that
λx
H
x = λ

x
H
x
58 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
Since x is an eigenvector, the scalar x
H
x is nonzero, so that we have
λ = λ

as promised. ∆
Corollary 5.1.3 A real and symmetric matrix has real eigenvalues.
Proof. A real and symmetric matrix is Hermitian. ∆
Theorem 5.1.4 Eigenvectors corresponding to distinct eigenvalues of a Hermitian matrix are mutually orthogonal.
Proof. Let λ and µ be two distinct eigenvalues of A, and let x and y be corresponding eigenvectors:
Ax = λx
Ay = µy ⇒ y
H
A = µy
H
because A = A
H
and from theorem 5.1.2 µ = µ

. If we multiply these two equations by y
H
from the left and x from
the right, respectively, we obtain
y
H
Ax = λy
H
x
y
H
Ax = µy
H
x ,
which implies
λy
H
x = µy
H
x
or
(λ −µ)y
H
x = 0 .
Since the two eigenvalues are distinct, λ −µ is nonzero, and we must have y
H
x = 0. ∆
Corollary 5.1.5 An n ×n Hermitian matrix with n distinct eigenvalues admits n orthonormal eigenvectors.
Proof. From theorem 5.1.4, the eigenvectors of an n×n Hermitian matrix with n distinct eigenvalues are all mutu-
ally orthogonal. Since the eigenvalue equation Ax = λx is homogeneous in x, the vector x can be normalized without
violating the equation. Consequently, the eigenvectors can be made to be orthonormal. ∆
In summary, any square matrix with n distinct eigenvalues can be diagonalized by a similarity transformation, and
any square Hermitian matrix with n distinct eigenvalues can be diagonalized by a unitary similarity transformation.
Notice that the converse is not true: a matrix can have coincident eigenvalues and still admit n independent, and
even orthonormal, eigenvectors. For instance, the n × n identity matrix has n equal eigenvalues but n orthonormal
eigenvectors (which can be chosen in infinitely many ways).
The examples in section 5.2 show that when some eigenvalues coincide, rather diverse situations can arise concern-
ing the eigenvectors. First, however, we point out a simple but fundamental fact about the eigenvalues of a triangular
matrix.
Theorem 5.1.6 The determinant of a triangular matrix is the product of the elements on its diagonal.
5.2. GOOD AND BAD MATRICES 59
Proof. This follows immediately from the definition of determinant. Without loss of generality, we can assume a
triangular matrix B to be upper-triangular, for otherwise we can repeat the argument for the transpose, which because
of the properties above has the same eigenvalues. Then, the only possibly nonzero b
i1
of the matrix B is b
11
, and the
summation in the definition of determinant given above reduces to a single term:
det(B) =
_
b
11
if B is 1 ×1
b
11
det(B
11
) otherwise
.
By repeating the argument for B
11
and so forth until we are left with a single scalar, we obtain
det(B) = b
11
· . . . · b
nn
.

Corollary 5.1.7 The eigenvalues of a triangular matrix are the elements on its diagonal.
Proof. The eigenvalues of a matrix A are the solutions of the equation
det(A−λI) = 0 .
If A is triangular, so is B = A−λI, and from the previous theorem we obtain
det(A−λI) = (a
11
−λ) · . . . · (a
nn
−λ)
which is equal to zero for
λ = a
11
, . . . , a
nn
.

Note that diagonal matrices are triangular, so this result holds for diagonal matrices as well.
5.2 Good and Bad Matrices
Solving differential equations becomes much easier when matrices have a full set of orthonormal eigenvectors. For
instance, the matrix
A =
_
2 0
0 1
_
(5.11)
has eigenvalues 2 and 1 and eigenvectors
s
1
=
_
1
0
_
s
2
=
_
0
1
_
.
Matrices with n orthonormal eigenvectors are called normal. Normal matrices are good news, because then the
n ×n system of differential equations
˙ x = Ax
has solution
x(t) =
n

i=1
c
i
s
i
e
λit
= S
_
¸
_
e
λ
1
t
.
.
.
e
λnt
_
¸
_c
60 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
where S = [s
1
· · · s
n
] are the eigenvectors, λ
i
are the eigenvalues, and the vector c of constants c
i
is
c = S
H
x(0) .
More compactly,
x(t) = S
_
¸
_
e
λ1t
.
.
.
e
λ
n
t
_
¸
_S
H
x(0) .
Fortunately these matrices occur frequently in practice. However, not all matrices are as good as these. First, there
may still be a complete set of n eigenvectors, but they may not be orthonormal. An example of such a matrix is
_
2 −1
0 1
_
which has eigenvalues 2 and 1 and nonorthogonal eigenvectors
q
1
=
_
1
0
_
q
2
=

2
2
_
1
1
_
.
This is conceptually only a slight problem, because the unitary matrix S is replaced by an invertible matrix Q, and the
solution becomes
x(t) = Q
_
¸
_
e
λ1t
.
.
.
e
λnt
_
¸
_Q
−1
x(0) .
Computationally this is more expensive, because a computation of a Hermitian is replaced by a matrix inversion.
However, things can be worse yet, and a full set of eigenvectors may fail to exist, as we now show.
A necessary condition for an n ×n matrix to be defective, that is, to have fewer than n eigenvectors, is that it have
repeated eigenvalues. In fact, we have seen (theorem 5.1.1) that a matrix with distinct eigenvalues (zero or nonzero
does not matter) has a full set of eigenvectors (perhaps nonorthogonal, but independent). The simplest example of a
defective matrix is
_
0 1
0 0
_
which has double eigenvalue 0 and only eigenvector [1 0]
T
, while
_
3 1
0 3
_
has double eigenvalue 3 and only eigenvector [1 0]
T
, so zero eigenvalues are not the problem.
However, repeated eigenvalues are not a sufficient condition for defectiveness, as the identity matrix proves.
How bad can a matrix be? Here is a matrix that is singular, has fewer than n eigenvectors, and the eigenvectors it
has are not orthogonal. It belongs to the scum of all matrices:
A =
_
_
0 2 −1
0 2 1
0 0 2
_
_
.
Its eigenvalues are 0, because the matrix is singular, and 2, repeated twice. A has to have a repeated eigenvalue if it is
to be defective. Its two eigenvectors are
q
1
=
_
_
1
0
0
_
_
, q
2
=

2
2
_
_
1
1
0
_
_
corresponding to eigenvalues 0 and 2 in this order, and there is no q
3
. Furthermore, q
1
and q
2
are not orthogonal to
each other.
5.3. COMPUTING EIGENVALUES AND EIGENVECTORS NUMERICALLY 61
5.3 Computing Eigenvalues and Eigenvectors Numerically
The examples above have shown that not every n × n matrix admits n independent eigenvectors, so some matrices
cannot be diagonalized by similarity transformations. Fortunately, these matrices can be triangularized by similarity
transformations, as we now show. We will show later on that this allows solving systems of linear differential equations
regardless of the structure of the system’s matrix of coefficients.
It is important to notice that if a matrix A is triangularized by similarity transformations,
T = Q
−1
AQ ,
then the eigenvalues of the triangular matrix T are equal to those of the original matrix A. In fact, if
Ax = λx ,
then
QTQ
−1
x = λx ,
that is,
Ty = λy
where
y = Q
−1
x ,
so λ is also an eigenvalue for T. The eigenvectors, however, are changed according to the last equation.
The Schur decomposition does even better, since it triangularizes any square matrix A by a unitary (possibly
complex) transformation:
T = S
H
AS .
This transformation is equivalent to factoring A into the product
A = STS
H
,
and this product is called the Schur decomposition of A. Numerically stable and efficient algorithms exist for the Schur
decomposition. In this note, we will not study these algorithms, but only show that all square matrices admit a Schur
decomposition.
5.3.1 Rotations into the x
1
Axis
An important preliminary fact concerns vector rotations. Let e
1
be the first column of the identity matrix. It is
intuitively obvious that any nonzero real vector x can be rotated into a vector parallel to e
1
. Formally, take any
orthogonal matrix S whose first column is
s
1
=
x
x
.
Since s
T
1
x = x
T
x/x = x, and since all the other s
j
are orthogonal to s
1
, we have
S
T
x =
_
¸
_
s
T
1
.
.
.
s
T
n
_
¸
_x =
_
¸
_
s
T
1
x
.
.
.
s
T
n
x
_
¸
_ =
_
¸
¸
¸
_
x
0
.
.
.
0
_
¸
¸
¸
_
which is parallel to e
1
as desired. It may be less obvious that a complex vector x can be transformed into a real vector
parallel to e
1
by a unitary transformation. But the trick is the same: let
s
1
=
x
x
.
62 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
Now s
1
may be complex. We have s
H
1
x = x
H
x/x = x, and
S
H
x =
_
¸
_
s
H
1
.
.
.
s
H
n
_
¸
_x =
_
¸
_
s
H
1
x
.
.
.
s
H
n
x
_
¸
_ =
_
¸
¸
¸
_
x
0
.
.
.
0
_
¸
¸
¸
_
,
just about like before. We are now ready to triangularize an arbitrary square matrix A.
5.3.2 The Schur Decomposition
The Schur decomposition theorem is the cornerstone of eigenvalue computations. It states that any square matrix can
be triangularized by unitary transformations. The diagonal elements of a triangular matrix are its eigenvalues, and
unitary transformations preserve eigenvalues. Consequently, if the Schur decomposition of a matrix can be computed,
its eigenvalues can be determined. Moreover, as we will see later, a system of linear differential equations can be
solved regardless of the structure of the matrix of its coefficients.
Lemma 5.3.1 If A is an n ×n matrix and λ and x are an eigenvalue of A and its corresponding eigenvector,
Ax = λx (5.12)
then there is a transformation
T = U
H
AU
where U is a unitary, n ×n matrix, such that
T =
_
¸
¸
¸
_
λ
0
.
.
.
0
C
_
¸
¸
¸
_
.
Proof. Let U be a unitary transformation that transforms the (possibly complex) eigenvector x of A into a real
vector on the x
1
axis:
x = U
_
¸
¸
¸
_
r
0
.
.
.
0
_
¸
¸
¸
_
where r is the nonzero norm of x. By substituting this into (5.12) and rearranging we have
AU
_
¸
¸
¸
_
r
0
.
.
.
0
_
¸
¸
¸
_
= λU
_
¸
¸
¸
_
r
0
.
.
.
0
_
¸
¸
¸
_
U
H
AU
_
¸
¸
¸
_
r
0
.
.
.
0
_
¸
¸
¸
_
= λ
_
¸
¸
¸
_
r
0
.
.
.
0
_
¸
¸
¸
_
U
H
AU
_
¸
¸
¸
_
1
0
.
.
.
0
_
¸
¸
¸
_
= λ
_
¸
¸
¸
_
1
0
.
.
.
0
_
¸
¸
¸
_
5.3. COMPUTING EIGENVALUES AND EIGENVECTORS NUMERICALLY 63
T
_
¸
¸
¸
_
1
0
.
.
.
0
_
¸
¸
¸
_
=
_
¸
¸
¸
_
λ
0
.
.
.
0
_
¸
¸
¸
_
.
The last left-hand side is the first column of T, and the corresponding right-hand side is of the form required by the
lemma. ∆
Theorem 5.3.2 (Schur) If A is any n ×n matrix then there exists a unitary n ×n matrix S such that
S
H
AS = T
where T is triangular. Furthermore, S can be chosen so that the eigenvalues λ
i
of A appear in any order along the
diagonal of T.
Proof. By induction. The theorem obviously holds for n = 1:
1 A1 = A .
Suppose it holds for all matrices of order n −1. Then from the lemma there exists a unitary U such that
U
H
AU =
_
¸
¸
¸
_
λ
0
.
.
.
0
C
_
¸
¸
¸
_
where λ is any eigenvalue of A. Partition C into a row vector and an (n −1) ×(n −1) matrix G:
C =
_
w
H
G
_
.
By the inductive hypothesis, there is a unitary matrix V such that V
H
GV is a Schur decomposition of G. Let
S = U
_
¸
¸
¸
_
1 0 · · · 0
0
.
.
.
0
V
_
¸
¸
¸
_
.
Clearly, S is a unitary matrix, and S
H
AS is upper-triangular. Since the elements on the diagonal of a triangular matrix
are the eigenvalues, S
H
AS is the Schur decomposition of A. Because we can pick any eigenvalue as λ, the order of
eigenvalues can be chosen arbitrarily. ∆
This theorem does not say how to compute the Schur decomposition, only that it exists. Fortunately, there is a
stable and efficient algorithm to compute the Schur decomposition. This is the preferred way to compute eigenvalues
numerically.
64 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
5.4 Eigenvalues/Vectors and Singular Values/Vectors
In this section we prove a few additional important properties of eigenvalues and eigenvectors. In the process, we also
establish a link between singular values/vectors and eigenvalues/vectors. While this link is very important, it is useful
to remember that eigenvalues/vectors and singular values/vectors are conceptually and factually very distinct entities
(recall figure 5.1).
First, a general relation between determinant and eigenvalues.
Theorem 5.4.1 The determinant of a matrix is equal to the product of its eigenvalues.
Proof. The proof is very simple, given the Schur decomposition. In fact, we know that the eigenvalues of a matrix
A are equal to those of the triangular matrix in the Schur decomposition of A. Furthermore, we know from theorem
5.1.6 that the determinant of a triangular matrix is the product of the elements on its diagonal. If we recall that a
unitary matrix has determinant 1 or -1, that the determinants of S and S
H
are the same, and that the determinant of a
product of matrices is equal to the product of the determinants, the proof is complete. ∆
We saw that an n ×n Hermitian matrix with n distinct eigenvalues admits n orthonormal eigenvectors (corollary
5.1.5). The assumption of distinct eigenvalues made the proof simple, but is otherwise unnecessary. In fact, now that
we have the Schur decomposition, we can state the following stronger result.
Theorem 5.4.2 (Spectral theorem) Every Hermitian matrix can be diagonalized by a unitary matrix, and every real
symmetric matrix can be diagonalized by an orthogonal matrix:
A = A
H
⇒ A = SΛS
H
A real, A = A
T
⇒ A = SΛS
T
, S real .
In either case, Λ is real and diagonal.
Proof. We already know that Hermitian matrices (and therefore real and symmetric ones) have real eigenvalues
(theorem 5.1.2), so Λ is real. Let now
A = STS
H
be the Schur decomposition of A. Since A is Hermitian, so is T. In fact, T = S
H
AS, and
T
H
= (S
H
AS)
H
= S
H
A
H
S = S
H
AS = T .
But the only way that T can be both triangular and Hermitian is for it to be diagonal, because 0

= 0. Thus, the
Schur decomposition of a Hermitian matrix is in fact a diagonalization, and this is the first equation of the theorem
(the diagonal of a Hermitian matrix must be real).
Let now Abe real and symmetric. All that is left to prove is that then its eigenvectors are real. But eigenvectors are
the solution of the homogeneous system (5.6), which is both real and rank-deficient, and therefore admits nontrivial
real solutions. Thus, S is real, and S
H
= S
T
. ∆
In other words, a Hermitian matrix, real or not, with distinct eigenvalues or not, has real eigenvalues and n or-
thonormal eigenvectors. If in addition the matrix is real, so are its eigenvectors.
We recall that a real matrix A such that for every nonzero x we have x
T
Ax > 0 is said to be positive definite. It is
positive semidefinite if for every nonzero x we have x
T
Ax ≥ 0. Notice that a positive definite matrix is also positive
semidefinite. Positive definite or semidefinite matrices arise in the solution of overconstrained linear systems, because
A
T
A is positive semidefinite for every A (lemma 5.4.5). They also occur in geometry through the equation of an
ellipsoid,
x
T
Qx = 1
5.4. EIGENVALUES/VECTORS AND SINGULAR VALUES/VECTORS 65
in which Q is positive definite. In physics, positive definite matrices are associated to quadratic forms x
T
Qx that
represent energies or second-order moments of mass or force distributions. Their physical meaning makes them
positive definite, or at least positive semidefinite (for instance, energies cannot be negative). The following result
relates eigenvalues/vectors with singular values/vectors for positive semidefinite matrices.
Theorem 5.4.3 The eigenvalues of a real, symmetric, positive semidefinite matrix A are equal to its singular values.
The eigenvectors of A are also its singular vectors, both left and right.
Proof. From the previous theorem, A = SΛS
T
, where both Λ and S are real. Furthermore, the entries in Λ are
nonnegative. In fact, from
As
i
= λs
i
we obtain
s
T
i
As
i
= s
T
i
λs
i
= λs
T
i
s
i
= λs
i

2
= λ .
If A is positive semidefinite, then x
T
Ax ≥ 0 for any nonzero x, and in particular s
T
i
As
i
≥ 0, so that λ ≥ 0.
But
A = SΛS
T
with nonnegative diagonal entries in Λ is the singular value decomposition A = UΣV
T
of A with Σ = Λ and
U = V = S. Recall that the eigenvalues in the Schur decomposition can be arranged in any desired order along the
diagonal. ∆
Theorem 5.4.4 A real, symmetric matrix is positive semidefinite iff all its eigenvalues are nonnegative. It is positive
definite iff all its eigenvalues are positive.
Proof. Theorem 5.4.3 implies one of the two directions: If A is real, symmetric, and positive semidefinite, then its
eigenvalues are nonnegative. If the proof of that theorem is repeated with the strict inequality, we also obtain that if A
is real, symmetric, and positive definite, then its eigenvalues are positive.
Conversely, we show that if all eigenvalues λ of a real and symmetric matrix A are positive (nonnegative) then A
is positive definite (semidefinite). To this end, let x be any nonzero vector. Since real and symmetric matrices have n
orthonormal eigenvectors (theorem 5.4.2), we can use these eigenvectors s
1
, . . . , s
n
as an orthonormal basis for R
n
,
and write
x = c
1
s
1
+. . . +c
n
s
n
with
c
i
= x
T
s
i
.
But then
x
T
Ax = x
T
A(c
1
s
1
+. . . +c
n
s
n
) = x
T
(c
1
As
1
+. . . +c
n
As
n
)
= x
T
(c
1
λ
1
s
1
+. . . +c
n
λ
n
s
n
) = c
1
λ
1
x
T
s
1
+. . . +c
n
λ
n
x
T
s
n
= λ
1
c
2
1
+. . . +λ
n
c
2
n
> 0 (or ≥ 0)
because the λ
i
are positive (nonnegative) and not all c
i
can be zero. Since x
T
Ax > 0 (or ≥ 0) for every nonzero x, A
is positive definite (semidefinite). ∆
Theorem 5.4.3 establishes one connection between eigenvalues/vectors and singular values/vectors: for symmetric,
positive definite matrices, the concepts coincide. This result can be used to introduce a less direct link, but for arbitrary
matrices.
Lemma 5.4.5 A
T
A is positive semidefinite.
66 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
Proof. For any nonzero x we can write x
T
A
T
Ax = Ax
2
≥ 0. ∆
Theorem 5.4.6 The eigenvalues of A
T
A with m ≥ n are the squares of the singular values of A; the eigenvectors of
A
T
Aare the right singular vectors of A. Similarly, for m ≤ n, the eigenvalues of AA
T
are the squares of the singular
values of A, and the eigenvectors of AA
T
are the left singular vectors of A.
Proof. If m ≥ n and A = UΣV
T
is the SVD of A, we have
A
T
A = V ΣU
T
UΣV
T
= V Σ
2
V
T
which is in the required format to be a (diagonal) Schur decomposition with S = V and T = Λ = Σ
2
. Similarly, for
m ≤ n,
AA
T
= UΣV
T
V ΣU
T
= UΣ
2
U
T
is a Schur decomposition with S = U and T = Λ = Σ
2
. ∆
We have seen that important classes of matrices admit a full set of orthonormal eigenvectors. The theorem below
characterizes the class of all matrices with this property, that is, the class of all normal matrices. To prove the theorem,
we first need a lemma.
Lemma 5.4.7 If for an n ×n matrix B we have BB
H
= B
H
B, then for every i = 1, . . . , n, the norm of the i-th row
of B equals the norm of its i-th column.
Proof. From BB
H
= B
H
B we deduce
Bx
2
= x
H
B
H
Bx = x
H
BB
H
x = B
H
x
2
. (5.13)
If x = e
i
, the i-th column of the n × n identity matrix, Be
i
is the i-th column of B, and B
H
e
i
is the i-th column of
B
H
, which is the conjugate of the i-th row of B. Since conjugation does not change the norm of a vector, the equality
(5.13) implies that the i-th column of B has the same norm as the i-th row of B. ∆
Theorem 5.4.8 An n ×n matrix is normal if an only if it commutes with its Hermitian:
AA
H
= A
H
A .
Proof. Let A = STS
H
be the Schur decomposition of A. Then,
AA
H
= STS
H
ST
H
S
H
= STT
H
S
H
and A
H
A = ST
H
S
H
STS
H
= ST
H
TS
H
.
Because S is invertible (even unitary), we have AA
H
= A
H
A if and only if TT
H
= T
H
T.
However, a triangular matrix T for which TT
H
= T
H
T must be diagonal. In fact, from the lemma, the norm of
the i-th row of T is equal to the norm of its i-th column. Let i = 1. Then, the first column of T has norm |t
11
|. The
first row has first entry t
11
, so the only way that its norm can be |t
11
| is for all other entries in the first row to be zero.
We now proceed through i = 2, . . . , n, and reason similarly to conclude that T must be diagonal.
The converse is also obviously true: if T is diagonal, then TT
H
= T
H
T. Thus, AA
H
= A
H
A if and only if T is
diagonal, that is, if and only if A can be diagonalized by a unitary similarity transformation. This is the definition of a
normal matrix. ∆
5.4. EIGENVALUES/VECTORS AND SINGULAR VALUES/VECTORS 67
Corollary 5.4.9 A triangular, normal matrix must be diagonal.
Proof. We proved this in the proof of theorem 5.4.8. ∆
Checking that A
H
A = AA
H
is much easier than computing eigenvectors, so theorem 5.4.8 is a very useful
characterization of normal matrices. Notice that Hermitian (and therefore also real symmetric) matrices commute
trivially with their Hermitians, but so do, for instance, unitary (and therefore also real orthogonal) matrices:
UU
H
= U
H
U = I .
Thus, Hermitian, real symmetric, unitary, and orthogonal matrices are all normal.
68 CHAPTER 5. EIGENVALUES AND EIGENVECTORS
Chapter 6
Ordinary Differential Systems
In this chapter we use the theory developed in chapter 5 in order to solve systems of first-order linear differential
equations with constant coefficients. These systems have the following form:
˙ x = Ax + b(t) (6.1)
x(0) = x
0
(6.2)
where x = x(t) is an n-dimensional vector function of time t, the dot denotes differentiation, the coefficients a
ij
in
the n ×n matrix A are constant, and the vector function b(t) is a function of time. The equation (6.2), in which x
0
is
a known vector, defines the initial value of the solution.
First, we show that scalar differential equations of order greater than one can be reduced to systems of first-order
differential equations. Then, in section 6.2, we recall a general result for the solution of first-order differential systems
from the elementary theory of differential equations. In section 6.3, we make this result more specific by showing
that the solution to a homogeneous system is a linear combination of exponentials multiplied by polynomials in t.
This result is based on the Schur decomposition introduced in chapter 5, which is numerically preferable to the more
commonly used Jordan canonical form. Finally, in sections 6.4 and 6.5, we set up and solve a particular differential
system as an illustrative example.
6.1 Scalar Differential Equations of Order Higher than One
The first-order system (6.1) subsumes also the case of a scalar differential equation of order n, possibly greater than 1,
d
n
y
dt
n
+c
n−1
d
n−1
y
dt
n−1
+. . . +c
1
dy
dt
+c
0
y = b(t) . (6.3)
In fact, such an equation can be reduced to a first-order system of the form (6.1) by introducing the n-dimensional
vector
x =
_
¸
_
x
1
.
.
.
x
n
_
¸
_ =
_
¸
¸
¸
_
y
dy
dt
.
.
.
d
n−1
y
dt
n−1
_
¸
¸
¸
_
.
With this definition, we have
d
i
y
dt
i
= x
i+1
for i = 0, . . . , n −1
d
n
y
dt
n
=
dx
n
dt
,
69
70 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
and x satisfies the additional n −1 equations
x
i+1
=
dx
i
dt
(6.4)
for i = 1, . . . , n − 1. If we write the original system (6.3) together with the n − 1 differential equations (6.4), we
obtain the first-order system
˙ x = Ax + b(t)
where
A =
_
¸
¸
¸
¸
¸
_
0 1 0 · · · 0
0 0 1 · · · 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 · · · 1
−c
0
−c
1
−c
2
· · · −c
n−1
_
¸
¸
¸
¸
¸
_
is the so-called companion matrix of (6.3) and
b(t) =
_
¸
¸
¸
¸
¸
_
0
0
.
.
.
0
b(t)
_
¸
¸
¸
¸
¸
_
.
6.2 General Solution of a Linear Differential System
We know from the general theory of differential equations that a general solution of system (6.1) with initial condition
(6.2) is given by
x(t) = x
h
(t) + x
p
(t)
where x
h
(t) is the solution of the homogeneous system
˙ x = Ax
x(0) = x
0
and x
p
(t) is a particular solution of
˙ x = Ax + b(t)
x(0) = 0 .
The two solution components x
h
and x
p
can be written by means of the matrix exponential, introduced in the following.
For the scalar exponential e
λt
we can write a Taylor series expansion
e
λt
= 1 +
λt
1!
+
λ
2
t
2
2!
+· · · =

j=0
λ
j
t
j
j!
.
Usually
1
, in calculus classes, the exponential is introduced by other means, and the Taylor series expansion above is
proven as a property.
For matrices, the exponential e
Z
of a matrix Z ∈ R
n×n
is instead defined by the infinite series expansion
e
Z
= I +
Z
1!
+
Z
2
2!
+· · · =

j=0
Z
j
j!
.
1
Not always. In some treatments, the exponential is defined through its Taylor series.
6.2. GENERAL SOLUTION OF A LINEAR DIFFERENTIAL SYSTEM 71
Here I is the n ×n identity matrix, and the general term Z
j
/j! is simply the matrix Z raised to the jth power divided
by the scalar j!. It turns out that this infinite sum converges (to an n×n matrix which we write as e
Z
) for every matrix
Z. Substituting Z = At gives
e
At
= I +
At
1!
+
A
2
t
2
2!
+
A
3
t
3
3!
+· · · =

j=0
A
j
t
j
j!
. (6.5)
Differentiating both sides of (6.5) gives
de
At
dt
= A+
A
2
t
1!
+
A
3
t
2
2!
+· · ·
= A
_
I +
At
1!
+
A
2
t
2
2!
+· · ·
_
de
At
dt
= Ae
At
.
Thus, for any vector w, the function x
h
(t) = e
At
w satisfies the homogeneous differential system
˙ x
h
= Ax
h
.
By using the initial values (6.2) we obtain w = x
0
, and
x
h
(t) = e
At
x(0) (6.6)
is a solution to the differential system (6.1) with b(t) = 0 and initial values (6.2). It can be shown that this solution is
unique.
From the elementary theory of differential equations, we also know that a particular solution to the nonhomoge-
neous (b(t) = 0) equation (6.1) is given by
x
p
(t) =
_
t
0
e
A(t−s)
b(s) ds .
This is easily verified, since by differentiating this expression for x
p
we obtain
˙ x
p
= Ae
At
_
t
0
e
−As
b(s) ds +e
At
e
−At
b(t) = Ax
p
+ b(t) ,
so x
p
satisfies equation (6.1).
In summary, we have the following result.
The solution to
˙ x = Ax + b(t) (6.7)
with initial value
x(0) = x
0
(6.8)
is
x(t) = x
h
(t) + x
p
(t) (6.9)
where
x
h
(t) = e
At
x(0) (6.10)
and
x
p
(t) =
_
t
0
e
A(t−s)
b(s) ds . (6.11)
72 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
Since we now have a formula for the general solution to a linear differential system, we seem to have all we need.
However, we do not know how to compute the matrix exponential. The naive solution to use the definition (6.5)
requires too many terms for a good approximation. As we have done for the SVD and the Schur decomposition, we
will only point out that several methods exist for computing a matrix exponential, but we will not discuss how this is
done
2
. In a fundamental paper on the subject, Nineteen dubious ways to compute the exponential of a matrix (SIAM
Review, vol. 20, no. 4, pp. 801-36), Cleve Moler and Charles Van Loan discuss a large number of different methods,
pointing out that no one of them is appropriate for all situations. A full discussion of this matter is beyond the scope
of these notes.
When the matrix A is constant, as we currently assume, we can be much more specific about the structure of the
solution (6.9) of system (6.7), and particularly so about the solution x
h
(t) to the homogeneous part. Specifically, the
matrix exponential (6.10) can be written as a linear combination, with constant vector coefficients, of scalar expo-
nentials multiplied by polynomials. In the general theory of linear differential systems, this is shown via the Jordan
canonical form. However, in the paper cited above, Moler and Van Loan point out that the Jordan form cannot be
computed reliably, and small perturbations in the data can change the results dramatically. Fortunately, a similar result
can be found through the Schur decomposition introduced in chapter 5. The next section shows how to do this.
6.3 Structure of the Solution
For the homogeneous case b(t) = 0, consider the first order system of linear differential equations
˙ x = Ax (6.12)
x(0) = x
0
. (6.13)
Two cases arise: either A admits n distinct eigenvalues, or is does not. In chapter 5, we have seen that if (but not only
if) A has n distinct eigenvalues then it has n linearly independent eigenvectors (theorem 5.1.1), and we have shown
how to find x
h
(t) by solving an eigenvalue problem. In section 6.3.1, we briefly review this solution. Then, in section
6.3.2, we show how to compute the homogeneous solution x
h
(t) in the extreme case of an n × n matrix A with n
coincident eigenvalues.
To be sure, we have seen that matrices with coincident eigenvalues can still have a full set of linearly independent
eigenvectors (see for instance the identity matrix). However, the solution procedure we introduce in section 6.3.2 for
the case of n coincident eigenvalues can be applied regardless to how many linearly independent eigenvectors exist.
If the matrix has a full complement of eigenvectors, the solution obtained in section 6.3.2 is the same as would be
obtained with the method of section 6.3.1.
Once these two extreme cases (nondefective matrix or all-coincident eigenvalues) have been handled, we show a
general procedure in section 6.3.3 for solving a homogeneous or nonhomogeneous differential system for any, square,
constant matrix A, defective or not. This procedure is based on backsubstitution, and produces a result analogous
to that obtained via Jordan decomposition for the homogeneous part x
h
(t) of the solution. However, since it is
based on the numerically sound Schur decomposition, the method of section 6.3.3 is superior in practice. For a
nonhomogeneous system, the procedure can be carried out analytically if the functions in the right-hand side vector
b(t) can be integrated.
6.3.1 A is Not Defective
In chapter 5 we saw how to find the homogeneous part x
h
(t) of the solution when A has a full set of n linearly
independent eigenvectors. This result is briefly reviewed in this section for convenience.
3
If A is not defective, then it has n linearly independent eigenvectors q
1
, . . . , q
n
with corresponding eigenvalues
λ
1
, . . . , λ
n
. Let
Q =
_
q
1
· · · q
n
¸
.
2
In Matlab, expm(A) is the matrix exponential of A.
3
Parts of this subsection and of the following one are based on notes written by Scott Cohen.
6.3. STRUCTURE OF THE SOLUTION 73
This square matrix is invertible because its columns are linearly independent. Since Aq
i
= λ
i
q
i
, we have
AQ = QΛ, (6.14)
where Λ = diag(λ
1
, . . . , λ
n
) is a square diagonal matrix with the eigenvalues of A on its diagonal. Multiplying both
sides of (6.14) by Q
−1
on the right, we obtain
A = QΛQ
−1
. (6.15)
Then, system (6.12) can be rewritten as follows:
˙ x = Ax
˙ x = QΛQ
−1
x
Q
−1
˙ x = ΛQ
−1
x
˙ y = Λy, (6.16)
where y = Q
−1
x. The last equation (6.16) represents n uncoupled, homogeneous, differential equations ˙ y
i
= λ
i
y
i
.
The solution is
y
h
(t) = e
Λt
y(0),
where
e
Λt
= diag(e
λ1t
, . . . , e
λnt
).
Using the relation x = Qy, and the consequent relation y(0) = Q
−1
x(0), we see that the solution to the homogeneous
system (6.12) is
x
h
(t) = Qe
Λt
Q
−1
x(0).
If A is normal, that is, if it has n orthonormal eigenvectors q
1
, . . . q
n
, then Q is replaced by the Hermitian matrix
S =
_
s
1
· · · s
n
¸
, Q
−1
is replaced by S
H
, and the solution to (6.12) becomes
x
h
(t) = Se
Λt
S
H
x(0).
6.3.2 A Has n Coincident Eigenvalues
When A = QΛQ
−1
, we derived that the solution to (6.12) is x
h
(t) = Qe
Λt
Q
−1
x(0). Comparing with (6.6), it should
be the case that
e
Q(Λt)Q
−1
= Qe
Λt
Q
−1
.
This follows easily from the definition of e
Z
and the fact that (Q(Λt)Q
−1
)
j
= Q(Λt)
j
Q
−1
. Similarly, if A = SΛS
H
,
where S is Hermitian, then the solution to (6.12) is x
h
(t) = Se
Λt
S
H
x(0), and
e
S(Λt)S
H
= Se
Λt
S
H
.
How can we compute the matrix exponential in the extreme case in which A has n coincident eigenvalues, regard-
less of the number of its linearly independent eigenvectors? In any case, A admits a Schur decomposition
A = STS
H
(theorem 5.3.2). We recall that S is a unitary matrix and T is upper triangular with the eigenvalues of Aon its diagonal.
Thus we can write T as
T = Λ +N,
where Λ is diagonal and N is strictly upper triangular. The solution (6.6) in this case becomes
x
h
(t) = e
S(Tt)S
H
x(0) = Se
Tt
S
H
x(0) = Se
Λt+Nt
S
H
x(0).
74 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
Thus we can compute (6.6) if we can compute e
Tt
= e
Λt+Nt
. This turns out to be almost as easy as computing e
Λt
when the diagonal matrix Λ is a multiple of the identity matrix:
Λ = λI
that is, when all the eigenvalues of A coincide. In fact, in this case, Λt and Nt commute:
Λt Nt = λIt Nt = λt Nt = Nt λt = Nt λIt = Nt Λt .
It can be shown that if two matrices Z
1
and Z
2
commute, that is if
Z
1
Z
2
= Z
2
Z
1
,
then
e
Z1+Z2
= e
Z1
e
Z2
= e
Z2
e
Z1
.
Thus, in our case, we can write
e
Λt+Nt
= e
Λt
e
Nt
.
We already know how to compute e
Λt
, so it remains to show how to compute e
Nt
. The fact that Nt is strictly upper
triangular makes the computation of this matrix exponential much simpler than for a general matrix Z.
Suppose, for example, that N is 4 ×4. Then N has three nonzero superdiagonals, N
2
has two nonzero superdiag-
onals, N
3
has one nonzero superdiagonal, and N
4
is the zero matrix:
N =
_
¸
¸
_
0 ∗ ∗ ∗
0 0 ∗ ∗
0 0 0 ∗
0 0 0 0
_
¸
¸
_
→N
2
=
_
¸
¸
_
0 0 ∗ ∗
0 0 0 ∗
0 0 0 0
0 0 0 0
_
¸
¸
_

N
3
=
_
¸
¸
_
0 0 0 ∗
0 0 0 0
0 0 0 0
0 0 0 0
_
¸
¸
_
→N
4
=
_
¸
¸
_
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
_
¸
¸
_
.
In general, for a strictly upper triangular n ×n matrix, we have N
j
= 0 for all j ≥ n (i.e., N is nilpotent of order n).
Therefore,
e
Nt
=

j=0
N
j
t
j
j!
=
n−1

j=0
N
j
t
j
j!
is simply a finite sum, and the exponential reduces to a matrix polynomial.
In summary, the general solution to the homogeneous differential system (6.12) with initial value (6.13) when the
n ×n matrix A has n coincident eigenvalues is given by
x
h
(t) = Se
Λt
n−1

j=0
N
j
t
j
j!
S
H
x
0
(6.17)
where
A = S(Λ +N)S
H
is the Schur decomposition of A,
Λ = λI
is a multiple of the identity matrix containing the coincident eigenvalues of A on its diagonal, and N is strictly upper
triangular.
6.3. STRUCTURE OF THE SOLUTION 75
6.3.3 The General Case
We are now ready to solve the linear differential system
˙ x = Ax + b(t) (6.18)
x(0) = x
0
(6.19)
in the general case of a constant matrix A, defective or not, with arbitrary b(t). In fact, let A = STS
H
be the Schur
decomposition of A, and consider the transformed system
˙ y(t) = Ty(t) + c(t) (6.20)
where
y(t) = S
H
x(t) and c(t) = S
H
b(t) . (6.21)
The triangular matrix T can always be written in the following form:
T =
_
¸
¸
¸
_
T
11
· · · · · · T
1k
0 T
22
· · · T
2k
.
.
.
.
.
.
.
.
.
0 · · · 0 T
kk
_
¸
¸
¸
_
where the diagonal blocks T
ii
for i = 1, . . . , k are of size n
i
×n
i
(possibly 1×1) and contain all-coincident eigenvalues.
The remaining nonzero blocks T
ij
with i < j can be in turn bundled into matrices
R
i
=
_
T
i,i+1
· · · T
i,k
¸
that contain everything to the right of the corresponding T
ii
. The vector c(t) can be partitioned correspondingly as
follows
c(t) =
_
¸
_
c
1
(t)
.
.
.
c
k
(t)
_
¸
_
where c
i
has n
i
entries, and the same can be done for
y(t) =
_
¸
_
y
1
(t)
.
.
.
y
k
(t)
_
¸
_
and for the initial values
y(0) =
_
¸
_
y
1
(0)
.
.
.
y
k
(0)
_
¸
_ .
The triangular system (6.20) can then be solved by backsubstitution as follows:
for i = k down to 1
if i < k
d
i
(t) = R
i
[y
i+1
(t), . . . , y
k
(t)]
T
else
d
i
(t) = 0 (an n
k
-dimensional vector of zeros)
end
T
ii
= λ
i
I +N
i
(diagonal and strictly upper-triangular part of T
ii
)
y
i
(t) = e
λ
i
It

n
i
−1
j=0
N
j
i
t
j
j!
y
i
(0) +
_
t
0
_
e
λ
i
I(t−s)

n
i
−1
j=0
N
j
i
(t−s)
j
j!
_
(c
i
(s) + d
i
(s)) ds
end.
76 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
In this procedure, the expression for y
i
(t) is a direct application of equations (6.9), (6.10), (6.11), and (6.17) with
S = I. In the general case, the applicability of this routine depends on whether the integral in the expression for y
i
(t)
can be computed analytically. This is certainly the case when b(t) is a constant vector b, because then the integrand is
a linear combination of exponentials multiplied by polynomials in t −s, which can be integrated by parts.
The solution x(t) for the original system (6.18) is then
x(t) = Sy(t) .
As an illustration, we consider a very small example, the 2 ×2 homogeneous, triangular case,
_
˙ y
1
˙ y
2
_
=
_
t
11
t
12
0 t
22
_ _
y
1
y
2
_
. (6.22)
When t
11
= t
22
= λ, we obtain
y(t) = e
λIt
_
1 t
12
t
0 1
_
y(0) .
In scalar form, this becomes
y
1
(t) = (y
1
(0) +t
12
y
2
(0) t) e
λt
y
2
(t) = y
2
(0) e
λt
,
and it is easy to verify that this solution satisfies the differential system (6.22).
When t
11
= λ
1
= t
22
= λ
2
, we could solve the system by finding the eigenvectors of T, since we know that in this
case two linearly independent eigenvectors exist (theorem 5.1.1). Instead, we apply the backsubstitution procedure
introduced in this section. The second equation of the system,
˙ y
2
(t) = t
22
y
2
has solution
y
2
(t) = y
2
(0) e
λ
2
t
.
We then have
d
1
(t) = t
12
y
2
(t) = t
12
y
2
(0) e
λ2t
and
y
1
(t) = y
1
(0)e
λ1t
+
_
t
0
e
λ1(t−s)
d
1
(s) ds
= y
1
(0)e
λ1t
+t
12
y
2
(0) e
λ1t
_
t
0
e
−λ1s
e
λ2s
ds
= y
1
(0)e
λ1t
+t
12
y
2
(0) e
λ1t
_
t
0
e
(λ2−λ1)s
ds
= y
1
(0)e
λ1t
+
t
12
y
2
(0)
λ
2
−λ
1
e
λ1t
(e
(λ2−λ1)t
−1)
= y
1
(0)e
λ
1
t
+
t
12
y
2
(0)
λ
2
−λ
1
(e
λ
2
t
−e
λ
1
t
)
Exercise: verify that this solution satisfies both the differential equation (6.22) and the initial value equation y(0) = y
0
.
Thus, the solutions to system (6.22) for t
11
= t
22
and for t
11
= t
22
have different forms. While y
2
(t) is the same
in both cases, we have
y
1
(t) = y
1
(0) e
λt
+t
12
y
2
(0) t e
λt
if t
11
= t
22
y
1
(t) = y
1
(0)e
λ1t
+
t
12
y
2
(0)
λ
2
−λ
1
(e
λ2t
−e
λ1t
) if t
11
= t
22
.
6.4. A CONCRETE EXAMPLE 77
rest position
of mass 1
rest position
of mass 2
2
v
1
v
1
2
2
1
3
Figure 6.1: A system of masses and springs. In the absence of external forces, the two masses would assume the
positions indicated by the dashed lines.
This would seem to present numerical difficulties when t
11
≈ t
22
, because the solution would suddenly switch
from one form to the other as the difference between t
11
and t
22
changes from about zero to exactly zero or viceversa.
This, however, is not a problem. In fact,
lim
λ
1
→λ
e
λt
−e
λ
1
t
λ −λ
1
= t e
λt
,
and the transition between the two cases is smooth.
6.4 A Concrete Example
In this section we set up and solve a more concrete example of a system of differential equations. The initial system
has two second-order equations, and is transformed into a first-order system with four equations. The 4 ×4 matrix of
the resulting system has an interesting structure, which allows finding eigenvalues and eigenvectors analytically with a
little trick. The point of this section is to show how to transform the complex formal solution of the differential system,
computed with any of the methods described above, into a real solution in a form appropriate to the problem at hand.
Consider the mechanical system in figure 6.1. Suppose that we want to study the evolution of the system over time.
Since forces are proportional to accelerations, because of Newton’s law, and since accelerations are second derivatives
of position, the new equations are differential. Because differentiation occurs only with respect to one variable, time,
these are ordinary differential equations, as opposed to partial.
In the following we write the differential equations that describe this system. Two linear differential equations of
the second order
4
result. We will then transform these into four linear differential equations of the first order.
By Hooke’s law, the three springs exert forces that are proportional to the springs’ elongations:
f
1
= c
1
v
1
f
2
= c
2
(v
2
−v
1
)
4
Recall that the order of a differential equation is the highest degree of derivative that appears in it.
78 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
f
3
= −c
3
v
2
where the c
i
are the positive spring constants (in newtons per meter).
The accelerations of masses 1 and 2 (springs are assumed to be massless) are proportional to their accelerations,
according to Newton’s second law:
m
1
¨ v
1
= −f
1
+f
2
= −c
1
v
1
+c
2
(v
2
−v
1
) = −(c
1
+c
2
)v
1
+c
2
v
2
m
2
¨ v
2
= −f
2
+f
3
= −c
2
(v
2
−v
1
) −c
3
v
2
= c
2
v
1
−(c
2
+c
3
)v
2
or, in matrix form,
¨v = Bv (6.23)
where
v =
_
v
1
v
2
_
and B =
_

c
1
+c
2
m
1
c
2
m
1
c
2
m
2

c
2
+c
3
m
2
_
.
We also assume that initial conditions
v(0) and ˙ v(0) (6.24)
are given, which specify positions and velocities of the two masses at time t = 0.
To solve the second-order system (6.23), we will first transform it to a system of four first-order equations. As
shown in the introduction to this chapter, the trick is to introduce variables to denote the first-order derivatives of v, so
that second-order derivatives of v are first-order derivatives of the new variables. For uniformity, we define four new
variables
u =
_
¸
¸
_
u
1
u
2
u
3
u
4
_
¸
¸
_
=
_
¸
¸
_
v
1
v
2
˙ v
1
˙ v
2
_
¸
¸
_
(6.25)
so that
u
3
= ˙ v
1
and u
4
= ˙ v
2
,
while the original system (6.23) becomes
_
˙ u
3
˙ u
4
_
= B
_
u
1
u
2
_
.
We can now gather these four first-order differential equations into a single system as follows:
˙ u = Au (6.26)
where
A =
_
¸
¸
_
0 0
0 0
1 0
0 1
B
0 0
0 0
_
¸
¸
_
.
Likewise, the initial conditions (6.24) are replaced by the (known) vector
u(0) =
_
v(0)
˙ v(0)
_
.
In the next section we solve equation (6.26).
6.5. SOLUTION OF THE EXAMPLE 79
6.5 Solution of the Example
Not all matrices have a full set of linearly independent eigenvectors. With the system of springs in figure 6.1, however,
we are lucky. The eigenvalues of A are solutions to the equation
Ax = λx , (6.27)
where we recall that
A =
_
0 I
B 0
_
and B =
_

c
1
+c
2
m1
c
2
m1
c
2
m2

c
2
+c
3
m2
_
.
Here, the zeros in A are 2 ×2 matrices of zeros, and I is the 2 ×2 identity matrix. If we partition the vector x into its
upper and lower halves y and z,
x =
_
y
z
_
,
we can write
Ax =
_
0 I
B 0
_ _
y
z
_
=
_
z
By
_
so that the eigenvalue equation (6.27) can be written as the following pair of equations:
z = λy (6.28)
By = λz ,
which yields
By = µy with µ = λ
2
.
In other words, the eigenvalues of A are the square roots of the eigenvalues of B: if we denote the two eigenvalues of
B as µ
1
and µ
2
, then the eigenvalues of A are
λ
1
=

µ
1
λ
2
= −

µ
1
λ
3
=

µ
2
λ
4
= −

µ
2
.
The eigenvalues µ
1
and µ
2
of B are the solutions of
det(B −µI) =
_
c
1
+c
2
m
1

__
c
2
+c
3
m
2

_

c
2
2
m
1
m
2
= µ
2
+ 2αµ +β = 0
where
α =
1
2
_
c
1
+c
2
m
1
+
c
2
+c
3
m
2
_
and β =
c
1
c
2
+c
1
c
3
+c
2
c
3
m
1
m
2
are positive constants that depend on the elastic properties of the springs and on the masses. We then obtain
µ
1,2
= −α ±γ ,
where
γ =
_
α
2
−β =
¸
1
4
_
c
1
+c
2
m
1

c
2
+c
3
m
2
_
2
+
c
2
2
m
1
m
2
.
The constant γ is real because the radicand is nonnegative. We also have that α ≥ γ, so that the two solutions µ
1,2
are
real and negative, and the four eigenvalues of A,
λ
1
=

−α +γ , λ
2
= −

−α +γ , (6.29)
λ
3
=

−α −γ , λ
4
= −

−α −γ (6.30)
80 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
come in nonreal, complex-conjugate pairs. This is to be expected, since our system of springs obviously exhibits an
oscillatory behavior.
Also the eigenvectors of A can be derived from those of B. In fact, from equation (6.28) we see that if y is an
eigenvector of B corresponding to eigenvalue µ = λ
2
, then there are two corresponding eigenvectors for A of the
form
x =
_
y
±λy
_
. (6.31)
The eigenvectors of B are the solutions of
(B −(−α ±γ)I)y = 0 . (6.32)
Since ±(−α±γ) are eigenvalues of B, the determinant of this equation is zero, and the two scalar equations in (6.32)
must be linearly dependent. The first equation reads

_
c
1
+c
2
m
1
−α ±γ
_
y
1
+
c
2
m
1
y
2
= 0
and is obviously satisfied by any vector of the form
y = k
_
c
2
m
1
c
1
+c
2
m
1
−α ±γ
_
where k is an arbitrary constant. For k = 0, y denotes the two eigenvectors of B, and from equation (6.31) the four
eigenvectors of A are proportional to the four columns of the following matrix:
Q =
_
¸
¸
_
c2
m1
c2
m1
c2
m1
c2
m1
a +λ
2
1
a +λ
2
2
a +λ
2
3
a +λ
2
4
λ
1
c2
m
1
λ
2
c2
m
1
λ
3
c2
m
1
λ
4
c2
m
1
λ
1
_
a +λ
2
1
_
λ
2
_
a +λ
2
2
_
λ
3
_
a +λ
2
3
_
λ
4
_
a +λ
2
4
_
_
¸
¸
_
(6.33)
where
a =
c
1
+c
2
m
1
.
The general solution to the first-order differential system (6.26) is then given by equation (6.17). Since we just found
four distinct eigenvectors, however, we can write more simply
u(t) = Qe
Λt
Q
−1
u(0) (6.34)
where
Λ =
_
¸
¸
_
λ
1
0 0 0
0 λ
2
0 0
0 0 λ
3
0
0 0 0 λ
4
_
¸
¸
_
.
In these expressions, the values of λ
i
are given in equations (6.30), and Q is in equation (6.33).
Finally, the solution to the original, second-order system (6.23) can be obtained from equation (6.25) by noticing
that v is equal to the first two components of u.
This completes the solution of our system of differential equations. However, it may be useful to add some
algebraic manipulation in order to show that the solution is indeed oscillatory. As we see in the following, the masses’
motions can be described by the superposition of two sinusoids whose frequencies depend on the physical constants
involved (masses and spring constants). The amplitudes and phases of the sinusoids, on the other hand, depend on the
initial conditions.
To simplify our manipulation, we note that
u(t) = Qe
Λt
w ,
6.5. SOLUTION OF THE EXAMPLE 81
where we defined
w = Q
−1
u(0) . (6.35)
We now leave the constants in w unspecified, and derive the general solution v(t) for the original, second-order
problem. Numerical values for the constants can be found from the initial conditions u(0) by equation (6.35). We
have
v(t) = Q(1 : 2, :)e
Λt
w ,
where Q(1 : 2, :) denotes the first two rows of Q. Since
λ
2
= −λ
1
and λ
4
= −λ
3
(see equations (6.30)), we have
Q(1 : 2, :) =
_
q
1
q
1
q
2
q
2
¸
where we defined
q
1
=
_
c
2
m1
c
1
+c
2
m1

2
1
_
and q
2
=
_
c
2
m1
c
1
+c
2
m1

2
3
_
.
Thus, we can write
v(t) = q
1
_
k
1
e
λ
1
t
+k
2
e
−λ
1
t
_
+ q
2
_
k
3
e
λ
3
t
+k
4
e
−λ
3
t
_
.
Since the λs are imaginary but v(t) is real, the k
i
must come in complex-conjugate pairs:
k
1
= k

2
and k
3
= k

4
. (6.36)
In fact, we have
v(0) = q
1
(k
1
+k
2
) + q
2
(k
3
+k
4
)
and from the derivative
˙ v(t) = q
1
λ
1
_
k
1
e
λ1t
−k
2
e
−λ1t
_
+ q
2
λ
3
_
k
3
e
λ3t
−k
4
e
−λ3t
_
we obtain
˙ v(0) = q
1
λ
1
(k
1
−k
2
) + q
2
λ
3
(k
3
−k
4
) .
Since the vectors q
i
are independent (assuming that the mass c
2
is nonzero), this means that
k
1
+k
2
is real k
1
−k
2
is purely imaginary
k
3
+k
4
is real k
3
−k
4
is purely imaginary ,
from which equations (6.36) follow.
Finally, by using the relation
e
jx
+e
−jx
2
= cos x ,
and simple trigonometry we obtain
v(t) = q
1
A
1
cos(ω
1
t +φ
1
) + q
2
A
2
cos(ω
2
t +φ
2
)
where
ω
1
=

α −γ =
¸
¸
¸
_
1
2
(a +b) −
¸
1
4
(a −b)
2
+
c
2
2
m
1
m
2
ω
2
=

α +γ =
¸
¸
¸
_
1
2
(a +b) +
¸
1
4
(a −b)
2
+
c
2
2
m
1
m
2
82 CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS
and
a =
c
1
+c
2
m
1
, b =
c
2
+c
3
m
2
.
Notice that these two frequencies depend only on the configuration of the system, and not on the initial conditions.
The amplitudes A
i
and phases φ
i
, on the other hand, depend on the constants k
i
as follows:
A
1
= 2|k
1
| , A
2
= 2|k
3
|
φ
1
= arctan
2
(Im(k
1
), Re(k
1
)) φ
2
= arctan
2
(Im(k
3
), Re(k
3
))
where Re, Im denote the real and imaginary part and where the two-argument function arctan
2
is defined as follows
for (x, y) = (0, 0)
arctan
2
(y, x) =
_
¸
¸
_
¸
¸
_
arctan(
y
x
) if x > 0
π + arctan(
y
x
) if x < 0
π
2
if x = 0 and y > 0

π
2
if x = 0 and y < 0
and is undefined for (x, y) = (0, 0). This function returns the arctangent of y/x (notice the order of the arguments) in
the proper quadrant, and extends the function by continuity along the y axis.
The two constants k
1
and k
3
can be found from the given initial conditions v(0) and ˙ v(0) from equations (6.35)
and (6.25).
Chapter 7
Stochastic State Estimation
Perhaps the most important part of studying a problem in robotics or vision, as well as in most other sciences, is to
determine a good model for the phenomena and events that are involved. For instance, studying manipulation requires
defining models for how a robot arm can move and for how it interacts with the world. Analyzing image motion
implies defining models for how points move in space and how this motion projects onto the image. When motion
is involved, as is very often the case, models take on frequently the form of dynamic systems. A dynamic system
is a mathematical description of a quantity that evolves over time. The theory of dynamic systems is both rich and
fascinating. Although in this chapter we will barely scratch its surface, we will consider one of its most popular and
useful aspects, the theory of state estimation, in the particular form of Kalman filtering. To this purpose, an informal
definition of a dynamic system is given in the next section. The definition is then illustrated by setting up the dynamic
system equations for a simple but realistic application, that of modeling the trajectory of an enemy mortar shell. In
sections 7.3 through 7.5, we will develop the theory of the Kalman filter, and in section 7.6 we will see that the shell
can be shot down before it hits us. As discussed in section 7.7, Kalman filtering has intimate connections with the
theory of algebraic linear systems we have developed in chapters 2 and 3.
7.1 Dynamic Systems
In its most general meaning, the term system refers to some physical entity on which some action is performed by
means of an input u. The system reacts to this input and produces an output y (see figure 7.1).
A dynamic system is a system whose phenomena occur over time. One often says that a system evolves over time.
Simple examples of a dynamic system are the following:
• An electric circuit, whose input is the current in a given branch and whose output is a voltage across a pair of
nodes.
• A chemical reactor, whose inputs are the external temperature, the temperature of the gas being supplied, and
the supply rate of the gas. The output can be the temperature of the reaction product.
• A mass suspended from a spring. The input is the force applied to the mass and the output is the position of the
mass.
input output system
S u y
Figure 7.1: A general system.
83
84 CHAPTER 7. STOCHASTIC STATE ESTIMATION
In all these examples, what is input and what is output is a choice that depends on the application. Also, all the
quantities in the examples vary continuously with time. In other cases, as for instance for switching networks and
computers, it is more natural to consider time as a discrete variable. If time varies continuously, the system is said to
be continuous; if time varies discretely, the system is said to be discrete.
7.1.1 State
Given a dynamic system, continuous or discrete, the modeling problem is to somehow correlate inputs (causes) with
outputs (effects). The examples above suggest that the output at time t cannot be determined in general by the value
assumed by the input quantity at the same point in time. Rather, the output is the result of the entire history of the
system. An effort of abstraction is therefore required, which leads to postulating a new quantity, called the state, which
summarizes information about the past and the present of the system. Specifically, the value x(t) taken by the state at
time t must be sufficient to determine the output at the same point in time. Also, knowledge of both x(t
1
) and u
[t
1
,t
2
)
,
that is, of the state at time t
1
and the input over the interval t
1
≤ t < t
2
, must allow computing the state (and hence
the output) at time t
2
. For the mass attached to a spring, for instance, the state could be the position and velocity of
the mass. In fact, the laws of classical mechanics allow computing the new position and velocity of the mass at time t
2
given its position and velocity at time t
1
and the forces applied over the interval [t
1
, t
2
). Furthermore, in this example,
the output y of the system happens to coincide with one of the two state variables, and is therefore always deducible
from the latter.
Thus, in a dynamic system the input affects the state, and the output is a function of the state. For a discrete
system, the way that the input changes the state at time instant number k into the new state at time instant k + 1 can
be represented by a simple equation:
x
k+1
= f(x
k
, u
k
, k)
where f is some function that represents the change, and u
k
is the input at time k. Similarly, the relation between state
and output can be expressed by another function:
y
k
= h(x
k
, k) .
A discrete dynamic system is completely described by these two equations and an initial state x
0
. In general, all
quantities are vectors.
For continuous systems, time does not come in quanta, so one cannot compute x
k+1
as a function of x
k
, u
k
, and
k, but rather compute x(t
2
) as a functional φ of x(t
1
) and the entire input u over the interval [t
1
, t
2
):
x(t
2
) = φ(x(t
1
), u(·), t
1
, t
2
)
where u(·) represents the entire function u, not just one of its values. A description of the system in terms of func-
tions, rather than functionals, can be given in the case of a regular system, for which the functional φ is continuous,
differentiable, and with continuous first derivative. In that case, one can show that there exists a function f such that
the state x(t) of the system satisfies the differential equation
˙ x(t) = f(x(t), u(t), t)
where the dot denotes differentiation with respect to time. The relation from state to output, on the other hand, is
essentially the same as for the discrete case:
y(t) = h(x(t), t) .
Specifying the initial state x
0
completes the definition of a continuous dynamic system.
7.1.2 Uncertainty
The systems defined in the previous section are called deterministic, since the evolution is exactly determined once
the initial state x at time 0 is known. Determinism implies that both the evolution function f and the output function
h are known exactly. This is, however, an unrealistic state of affairs. In practice, the laws that govern a given physical
7.2. AN EXAMPLE: THE MORTAR SHELL 85
system are known up to some uncertainty. In fact, the equations themselves are simple abstractions of a complex
reality. The coefficients that appear in the equations are known only approximately, and can change over time as a
result of temperature changes, component wear, and so forth. A more realistic model then allows for some inherent,
unresolvable uncertainty in both f and h. This uncertainty can be represented as noise that perturbs the equations we
have presented so far. A discrete system then takes on the following form:
x
k+1
= f(x
k
, u
k
, k) +η
k
y
k
= h(x
k
, k) +ξ
k
and for a continuous system
˙ x(t) = f(x(t), u(t), t) +η(t)
y(t) = h(x(t), t) +ξ(t) .
Without loss of generality, the noise distributions can be assumed to have zero mean, for otherwise the mean can be
incorporated into the deterministic part, that is, in either f or h. The mean may not be known, but this is a different
story: in general the parameters that enter into the definitions of f and h must be estimated by some method, and the
mean perturbations are no different.
A common assumption, which is sometimes valid and always simplifies the mathematics, is that η and ξ are
zero-mean Gaussian random variables with known covariance matrices Q and R, respectively.
7.1.3 Linearity
The mathematics becomes particularly simple when both the evolution function f and the output function h are linear.
Then, the system equations become
x
k+1
= F
k
x
k
+G
k
u
k

k
y
k
= H
k
x
k

k
for the discrete case, and
˙ x(t) = F(t)x(t) +G(t)u(t) +η(t)
y(t) = H(t)x(t) +ξ(t)
for the continuous one. It is useful to specify the sizes of the matrices involved. We assume that the input u is a vector
in R
p
, the state x is in R
n
, and the output y is in R
m
. Then, the state propagation matrix F is n ×n, the input matrix
G is n × p, and the output matrix H is m × n. The covariance matrix Q of the system noise η is n × n, and the
covariance matrix of the output noise ξ is m×m.
7.2 An Example: the Mortar Shell
In this section, the example of the mortar shell will be discussed in order to see some of the technical issues involved
in setting up the equations of a dynamic system. In particular, we consider discretization issues because the physical
system is itself continuous, but we choose to model it as a discrete system for easier implementation on a computer.
In sections 7.3 through 7.5, we consider the state estimation problem: given observations of the output y over an
interval of time, we want to determine the state x of the system. This is a very important task. For instance, in the case
of the mortar shell, the state is the (initially unknown) position and velocity of the shell, while the output is a set of
observations made by a tracking system. Estimating the state then leads to enough knowledge about the shell to allow
driving an antiaircraft gun to shoot the shell down in mid-flight.
You spotted an enemy mortar installation about thirty kilometers away, on a hill that looks about 0.5 kilometers
higher than your own position. You want to track incoming projectiles with a Kalman filter so you can aim your guns
86 CHAPTER 7. STOCHASTIC STATE ESTIMATION
accurately. You do not know the initial velocity of the projectiles, so you just guess some values: 0.6 kilometers/second
for the horizontal component, 0.1 kilometers/second for the vertical component. Thus, your estimate of the initial state
of the projectile is
ˆx
0
=
_
¸
¸
_
˙
d
d
˙ z
z
_
¸
¸
_
=
_
¸
¸
_
−0.6
30
0.1
0.5
_
¸
¸
_
where d is the horizontal coordinate, z is the vertical, you are at (0, 0), and dots denote derivatives with respect to
time.
From your high-school physics, you remember that the laws of motion for a ballistic trajectory are the following:
d(t) = d(0) +
˙
d(0)t (7.1)
z(t) = z(0) + ˙ z(0)t −
1
2
gt
2
(7.2)
where g is the gravitational acceleration, equal to 9.8 × 10
−3
kilometers per second squared. Since you do not trust
your physics much, and you have little time to get ready, you decide to ignore air drag. Because of this, you introduce
a state update covariance matrix Q = 0.1I
4
, where I
4
is the 4 ×4 identity matrix.
All you have to track the shells is a camera pointed at the mortar that will rotate so as to keep the projectile at the
center of the image, where you see a blob that increases in size as the projectile gets closer. Thus, the aiming angle of
the camera gives you elevation information about the projectile’s position, and the size of the blob tells you something
about the distance, given that you know the actual size of the projectiles used and all the camera parameters. The
projectile’s elevation is
e = 1000
z
d
(7.3)
when the projectile is at (d, z). Similarly, the size of the blob in pixels is
s =
1000

d
2
+z
2
. (7.4)
You do not have very precise estimates of the noise that corrupts e and s, so you guess measurement covariances
R
e
= R
s
= 1000, which you put along the diagonal of a 2 ×2 diagonal measurement covariance matrix R.
7.2.1 The Dynamic System Equation
Equations (7.1) and (7.2) are continuous. Since you are taking measurements every dt = 0.2 seconds, you want to
discretize these equations. For the z component, equation (7.2) yields
z(t +dt) −z(t) = z(0) + ˙ z(0)(t +dt) −
1
2
g(t +dt)
2

_
z(0) + ˙ z(0)t −
1
2
gt
2
_
= ( ˙ z(0) −gt)dt −
1
2
g(dt)
2
= ˙ z(t)dt −
1
2
g(dt)
2
,
since ˙ z(0) −gt = ˙ z(t).
Consequently, if t +dt is time instant k + 1 and t is time instant k, you have
z
k+1
= z
k
+ ˙ z
k
dt −
1
2
g(dt)
2
. (7.5)
The reasoning for the horizontal component d is the same, except that there is no acceleration:
d
k+1
= d
k
+
˙
d
k
dt . (7.6)
7.2. AN EXAMPLE: THE MORTAR SHELL 87
Equations (7.5) and (7.6) can be rewritten as a single system update equation
x
k+1
= Fx
k
+Gu
where
x
k
=
_
¸
¸
_
˙
d
k
d
k
˙ z
k
z
k
_
¸
¸
_
is the state, the 4 × 4 matrix F depends on dt, the control scalar u is equal to −g, and the 4 × 1 control matrix G
depends on dt. The two matrices F and G are as follows:
F =
_
¸
¸
_
1 0 0 0
dt 1 0 0
0 0 1 0
0 0 dt 1
_
¸
¸
_
G =
_
¸
¸
_
0
0
dt
−dt
2
2
_
¸
¸
_
.
7.2.2 The Measurement Equation
The two nonlinear equations (7.3) and (7.4) express the available measurements as a function of the true values of the
projectile coordinates d and z. We want to replace these equations with linear approximations. To this end, we develop
both equations as Taylor series around the current estimate and truncate them after the linear term. From the elevation
equation (7.3), we have
e
k
= 1000
z
k
d
k
≈ 1000
_
ˆ z
k
ˆ
d
k
+
z
k
− ˆ z
k
ˆ
d
k

ˆ z
k
ˆ
d
2
k
(d
k

ˆ
d
k
)
_
,
so that after simplifying we can redefine the measurement to be the discrepancy from the estimated value:
e

k
= e
k
−1000
ˆ z
k
ˆ
d
k
≈ 1000(
z
k
ˆ
d
k

ˆ z
k
ˆ
d
2
k
d
k
) . (7.7)
We can proceed similarly for equation (7.4):
s
k
=
1000
_
d
2
k
+z
2
k

1000
_
ˆ
d
2
k
+ ˆ z
2
k

1000
ˆ
d
k
(
ˆ
d
2
k
+ ˆ z
2
k
)
3/2
(d
k

ˆ
d
k
) −
1000ˆ z
k
(
ˆ
d
2
k
+ ˆ z
2
k
)
3/2
(z
k
− ˆ z
k
)
and after simplifying:
s

k
= s
k

2000
_
ˆ
d
2
+ ˆ z
2
≈ −1000
_
ˆ
d
k
(
ˆ
d
2
k
+ ˆ z
2
k
)
3/2
d
k
+
ˆ z
k
(
ˆ
d
2
k
+ ˆ z
2
k
)
3/2
z
k
_
. (7.8)
The two measurements s

k
and e

k
just defined can be collected into a single measurement vector
y
k
=
_
s

k
e

k
_
,
and the two approximate measurement equations (7.7) and (7.8) can be written in the matrix form
y
k
= H
k
x
k
(7.9)
where the measurement matrix H
k
depends on the current state estimate ˆx
k
:
H
k
= −1000
_
_
0
ˆ
d
k
(
ˆ
d
2
k
+ˆ z
2
k
)
3/2
0
ˆ z
k
(
ˆ
d
2
k
+ˆ z
2
k
)
3/2
0
ˆ z
k
ˆ
d
2
k
0 −
1
ˆ
d
k
_
_
88 CHAPTER 7. STOCHASTIC STATE ESTIMATION
As the shell approaches us, we frantically start studying state estimation, and in particular Kalman filtering, in the
hope to build a system that lets us shoot down the shell before it hits us. The next few sections will be read under this
impending threat.
Knowing the model for the mortar shell amounts to knowing the laws by which the object moves and those that
relate the position of the projectile to our observations. So what else is there left to do? From the observations, we
would like to know where the mortar shell is right now, and perhaps predict where it will be in a few seconds, so we
can direct an antiaircraft gun to shoot down the target. In other words, we want to know x
k
, the state of the dynamic
system. Clearly, knowing x
0
instead is equivalent, at least when the dynamics of the system are known exactly (the
system noise η
k
is zero). In fact, from x
0
we can simulate the system up until time t, thereby determining x
k
as well.
Most importantly, we do not want to have all the observations before we shoot: we would be dead by then. A scheme
that refines an initial estimation of the state as new observations are acquired is called a recursive
1
state estimation
system. The Kalman filter is one of the most versatile schemes for recursive state estimations. The original paper
by Kalman (R. E. Kalman, “A new approach to linear filtering and prediction problems,” Transactions of the ASME
Journal Basic Engineering, 82:34–45, 1960) is still one of the most readable treatments of this subject from the point
of view of stochastic estimation.
Even without noise, a single observation y
k
may not be sufficient to determine the state x
k
(in the example, one
observation happens to be sufficient). This is a very interesting aspect of state estimation. It is really the ensemble of
all observations that let one estimate the state, and yet observations are processed one at a time, as they become avail-
able. A classical example of this situation in computer vision is the reconstruction of three-dimensional shape from
a sequence of images. A single image is two-dimensional, so by itself it conveys no three-dimensional information.
Kalman filters exist that recover shape information froma sequence of images. See for instance L. Matthies, T. Kanade,
and R. Szeliski, “Kalman filter-based algorithms for estimating depth from image sequences,” International Journal of
Computer Vision, 3(3):209-236, September 1989; and T.J. Broida, S. Chandrashekhar, and R. Chellappa, “Recursive
3-D motion estimation from a monocular image sequence,” IEEE Transactions on Aerospace and Electronic Systems,
26(4):639–656, July 1990.
Here, we introduce the Kalman filter from the simpler point of view of least squares estimation, since we have
developed all the necessary tools in the first part of this course. The next section defines the state estimation problem
for a discrete dynamic system in more detail. Then, section 7.4 defines the essential notions of estimation theory
that are necessary to understand the quantitative aspects of Kalman filtering. Section 7.5 develops the equation of the
Kalman filter, and section 7.6 reconsiders the example of the mortar shell. Finally, section 7.7 establishes a connection
between the Kalman filter and the solution of a linear system.
7.3 State Estimation
In this section, the estimation problem is defined in some more detail. Given a discrete dynamic system
x
k+1
= F
k
x
k
+G
k
u
k

k
(7.10)
y
k
= H
k
x
k

k
(7.11)
where the system noise η
k
and the measurement noise ξ
k
are Gaussian variables,
η
k
∼ N(0, Q
k
)
ξ
k
∼ N(0, R
k
) ,
as well as a (possibly completely wrong) estimate ˆx
0
of the initial state and an initial covariance matrix P
0
of the
estimate ˆx
0
, the Kalman filter computes the optimal estimate ˆx
k|k
at time k given the measurements y
0
, . . . , y
k
. The
filter also computes an estimate P
k|k
of the covariance of ˆx
k|k
given those measurements. In these expressions, the hat
means that the quantity is an estimate. Also, the first k in the subscript refers to which variable is being estimated, the
second to which measurements are being used for the estimate. Thus, in general, ˆx
i|j
is the estimate of the value that
x assumes at time i given the first j + 1 measurements y
0
, . . . , y
j
.
1
The term “recursive” in the systems theory literature corresponds loosely to “incremental” or “iterative” in computer science.
7.3. STATE ESTIMATION 89
k | k-1
^
y
H
k
x
k | k-1
^
x
^
k | k
k | k
P
x
^
P
k+1 | k
k+1 | k
propagate propagate
x
^
P
k-1 | k-1
k-1 | k-1
y
k
y
k-1
y
k+1
k
update
k | k-1
P
k-1 k+1
time
Figure 7.2: The update stage of the Kalman filter changes the estimate of the current system state x
k
to make the
prediction of the measurement closer to the actual measurement y
k
. Propagation then accounts for the evolution of the
system state, as well as the consequent growing uncertainty.
7.3.1 Update
The covariance matrix P
k|k
must be computed in order to keep the Kalman filter running, in the following sense. At
time k, just before the new measurement y
k
comes in, we have an estimate ˆx
k|k−1
of the state vector x
k
based on the
previous measurements y
0
, . . . , y
k−1
. Now we face the problem of incorporating the new measurement y
k
into our
estimate, that is, of transforming ˆx
k|k−1
into ˆx
k|k
. If ˆx
k|k−1
were exact, we could compute the new measurement y
k
without even looking at it, through the measurement equation (7.11). Even if ˆx
k|k−1
is not exact, the estimate
ˆy
k|k−1
= H
k
ˆx
k|k−1
is still our best bet. Now y
k
becomes available, and we can consider the residue
r
k
= y
k
−ˆy
k|k−1
= y
k
−H
k
ˆx
k|k−1
.
If this residue is nonzero, we probably need to correct our estimate of the state x
k
, so that the new prediction
ˆy
k|k
= H
k
ˆx
k|k
of the measurement value is closer to the measurement y
k
than the old prediction
ˆy
k|k−1
= H
k
ˆx
k|k−1
that we made just before the new measurement y
k
was available.
The question however is, by how much should we correct our estimate of the state? We do not want to make ˆy
k|k
coincide with y
k
. That would mean that we trust the new measurement completely, but that we do not trust our state
estimate ˆx
k|k−1
at all, even if the latter was obtained through a large number of previous measurements. Thus, we
need some criterion for comparing the quality of the new measurement y
k
with that of our old estimate ˆx
k|k−1
of the
state. The uncertainty about the former is R
k
, the covariance of the observation error. The uncertainty about the state
just before the new measurement y
k
becomes available is P
k|k−1
. The update stage of the Kalman filter uses R
k
and
P
k|k−1
to weigh past evidence (ˆx
k|k−1
) and new observations (y
k
). This stage is represented graphically in the middle
of figure 7.2. At the same time, also the uncertainty measure P
k|k−1
must be updated, so that it becomes available for
the next step. Because a new measurement has been read, this uncertainty becomes usually smaller: P
k|k
< P
k|k−1
.
The idea is that as time goes by the uncertainty on the state decreases, while that about the measurements may
remain the same. Then, measurements count less and less as the estimate approaches its true value.
90 CHAPTER 7. STOCHASTIC STATE ESTIMATION
7.3.2 Propagation
Just after arrival of the measurement y
k
, both state estimate and state covariance matrix have been updated as described
above. But between time k and time k + 1 both state and covariance may change. The state changes according to the
system equation (7.10), so our estimate ˆx
k+1|k
of x
k+1
given y
0
, . . . , y
k
should reflect this change as well. Similarly,
because of the system noise η
k
, our uncertainty about this estimate may be somewhat greater than one time epoch ago.
The system equation (7.10) essentially “dead reckons” the new state from the old, and inaccuracies in our model of
how this happens lead to greater uncertainty. This increase in uncertainty depends on the system noise covariance Q
k
.
Thus, both state estimate and covariance must be propagated to the new time k + 1 to yield the new state estimate
ˆx
k+1|k
and the new covariance P
k+1|k
. Both these changes are shown on the right in figure 7.2.
In summary, just as the state vector x
k
represents all the information necessary to describe the evolution of a
deterministic system, the covariance matrix P
k|k
contains all the necessary information about the probabilistic part of
the system, that is, about how both the system noise η
k
and the measurement noise ξ
k
corrupt the quality of the state
estimate ˆx
k|k
.
Hopefully, this intuitive introduction to Kalman filtering gives you an idea of what the filter does, and what infor-
mation it needs to keep working. To turn these concepts into a quantitative algorithm we need some preliminaries on
optimal estimation, which are discussed in the next section. The Kalman filter itself is derived in section 7.5.
7.4 BLUE Estimators
In what sense does the Kalman filter use covariance information to produce better estimates of the state? As we will
se later, the Kalman filter computes the Best Linear Unbiased Estimate (BLUE) of the state. In this section, we see
what this means, starting with the definition of a linear estimation problem, and then considering the attributes “best”
and “unbiased” in turn.
7.4.1 Linear Estimation
Given a quantity y (the observation) that is a known function of another (deterministic but unknown) quantity x (the
state) plus some amount of noise,
y = h(x) + n , (7.12)
the estimation problem amounts to finding a function
ˆx = L(y)
such that ˆx is as close as possible to x. The function L is called an estimator, and its value ˆx given the observations y
is called an estimate. Inverting a function is an example of estimation. If the function h is invertible and the noise term
n is zero, then L is the inverse of h, no matter how the phrase “as close as possible” is interpreted. In fact, in that case
ˆx is equal to x, and any distance between ˆx and x must be zero. In particular, solving a square, nonsingular system
y = Hx (7.13)
is, in this somewhat trivial sense, a problem of estimation. The optimal estimator is then represented by the matrix
L = H
−1
and the optimal estimate is
ˆx = Ly .
A less trivial example occurs, for a linear observation function, when the matrix H has more rows than columns,
so that the system (7.13) is overconstrained. In this case, there is usually no inverse to H, and again one must say in
what sense ˆx is required to be “as close as possible” to x. For linear systems, we have so far considered the criterion
that prefers a particular ˆx if it makes the Euclidean norm of the vector y − Hx as small as possible. This is the
7.4. BLUE ESTIMATORS 91
(unweighted) least squares criterion. In section 7.4.2, we will see that in a very precise sense ordinary least squares
solve a particular type of estimation problem, namely, the estimation problem for the observation equation (7.12) with
h a linear function and n Gaussian zero-mean noise with the indentity matrix for covariance.
An estimator is said to be linear if the function L is linear. Notice that the observation function h can still be
nonlinear. If L is required to be linear but h is not, we will probably have an estimator that produces a worse estimate
than a nonlinear one. However, it still makes sense to look for the best possible linear estimator. The best estimator
for a linear observation function happens to be a linear estimator.
7.4.2 Best
In order to define what is meant by a “best” estimator, one needs to define a measure of goodness of an estimate. In
the least squares approach to solving a linear system like (7.13), this distance is defined as the Euclidean norm of the
residue vector
y −Hˆx
between the left and the right-hand sides of equation (7.13), evaluated at the solution ˆx. Replacing (7.13) by a “noisy
equation”,
y = Hx + n (7.14)
does not change the nature of the problem. Even equation (7.13) has no exact solution when there are more independent
equations than unknowns, so requiring equality is hopeless. What the least squares approach is really saying is that
even at the solution ˆx there is some residue
n = y −Hˆx (7.15)
and we would like to make that residue as small as possible in the sense of the Euclidean norm. Thus, an overcon-
strained system of the form (7.13) and its “noisy” version (7.14) are really the same problem. In fact, (7.14) is the
correct version, if the equality sign is to be taken literally.
The noise term, however, can be used to generalize the problem. In fact, the Euclidean norm of the residue (7.15)
treats all components (all equations in (7.14)) equally. In other words, each equation counts the same when computing
the norm of the residue. However, different equations can have noise terms of different variance. This amounts to
saying that we have reasons to prefer the quality of some equations over others or, alternatively, that we want to
enforce different equations to different degrees. From the point of view of least squares, this can be enforced by some
scaling of the entries of n or, even, by some linear transformation of them:
n →Wn
so instead of minimizing n
2
= n
T
n (the square is of course irrelevant when it comes to minimization), we now
minimize
Wn
2
= n
T
R
−1
n
where
R
−1
= W
T
W
is a symmetric, nonnegative-definite matrix. This minimization problem, called weighted least squares, is only slightly
different from its unweighted version. In fact, we have
Wn = W(y −Hx) = Wy −WHx
so we are simply solving the system
Wy = WHx
in the traditional, “unweighted” sense. We know the solution from normal equations:
ˆx = ((WH)
T
WH)
−1
(WH)
T
Wy = (H
T
R
−1
H)
−1
H
T
R
−1
y .
92 CHAPTER 7. STOCHASTIC STATE ESTIMATION
Interestingly, this same solution is obtained from a completely different criterion of goodness of a solution ˆx. This
criterion is a probabilistic one. We consider this different approach because it will let us show that the Kalman filter is
optimal in a very useful sense.
The new criterion is the so-called minimum-covariance criterion. The estimate ˆx of x is some function of the
measurements y, which in turn are corrupted by noise. Thus, ˆx is a function of a random vector (noise), and is therefore
a random vector itself. Intuitively, if we estimate the same quantity many times, from measurements corrupted by
different noise samples from the same distribution, we obtain different estimates. In this sense, the estimates are
random.
It makes therefore sense to measure the quality of an estimator by requiring that its variance be as small as possible:
the fluctuations of the estimate ˆx with respect to the true (unknown) value x from one estimation experiment to the
next should be as small as possible. Formally, we want to choose a linear estimator L such that the estimates ˆx = Ly
it produces minimize the following covariance matrix:
P = E[(x −ˆx)(x −ˆx)
T
] .
Minimizing a matrix, however, requires a notion of “size” for matrices: how large is P? Fortunately, most inter-
esting matrix norms are equivalent, in the sense that given two different definitions P
1
and P
2
of matrix norm
there exist two positive scalars α, β such that
αP
1
< P
2
< βP
1
.
Thus, we can pick any norm we like. In fact, in the derivations that follow, we only use properties shared by all norms,
so which norm we actually use is irrelevant. Some matrix norms were mentioned in section 3.2.
7.4.3 Unbiased
In additionto requiring our estimator to be linear and with minimum covariance, we also want it to be unbiased, in the
sense that if repeat the same estimation experiment many times we neither consistently overestimate nor consistently
underestimate x. Mathematically, this translates into the following requirement:
E[x −ˆx] = 0 and E[ˆx] = E[x] .
7.4.4 The BLUE
We now address the problem of finding the Best Linear Unbiased Estimator (BLUE)
ˆx = Ly
of x given that y depends on x according to the model (7.14), which is repeated here for convenience:
y = Hx + n . (7.16)
First, we give a necessary and sufficient condition for L to be unbiased.
Lemma 7.4.1 Let n in equation (7.16) be zero mean. Then the linear estimator L is unbiased if an only if
LH = I ,
the identity matrix.
Proof.
E[x −ˆx] = E[x −Ly] = E[x −L(Hx + n)]
= E[(I −LH)x] −E[Ln] = (I −HL)E[x]
7.4. BLUE ESTIMATORS 93
since E[Ln] = LE[n] and E[n] = 0. For this to hold for all x we need I −LH = 0. ∆
And now the main result.
Theorem 7.4.2 The Best Linear Unbiased Estimator (BLUE)
ˆx = Ly
for the measurement model
y = Hx + n
where the noise vector n has zero mean and covariance R is given by
L = (H
T
R
−1
H)
−1
H
T
R
−1
and the covariance of the estimate ˆx is
P = E[(x −ˆx)(x −ˆx)
T
] = (H
T
R
−1
H)
−1
. (7.17)
Proof. We can write
P = E[(x −ˆx)(x −ˆx)
T
] = E[(x −Ly)(x −Ly)
T
]
= E[(x −LHx −Ln)(x −LHx −Ln)
T
] = E[((I −LH)x −Ln)((I −LH)x −Ln)
T
]
= E[Lnn
T
L
T
] = LE[nn
T
] L
T
= LRL
T
because L is unbiased, so that LH = I.
To show that
L
0
= (H
T
R
−1
H)
−1
H
T
R
−1
(7.18)
is the best choice, let L be any (other) linear unbiased estimator. We can trivially write
L = L
0
+ (L −L
0
)
and
P = LRL
T
= [L
0
+ (L −L
0
)]R[L
0
+ (L −L
0
)]
T
= L
0
RL
T
0
+ (L −L
0
)RL
T
0
+L
0
R(L −L
0
)
T
+ (L −L
0
)R(L −L
0
)
T
.
From (7.18) we obtain
RL
T
0
= RR
−1
H(H
T
R
−1
H)
−1
= H(H
T
R
−1
H)
−1
so that
(L −L
0
)RL
T
0
= (L −L
0
)H(H
T
R
−1
H)
−1
= (LH −L
0
H)(H
T
R
−1
H)
−1
.
But L and L
0
are unbiased, so LH = L
0
H = I, and
(L −L
0
)RL
T
0
= 0 .
The term L
0
R(L −L
0
)
T
is the transpose of this, so it is zero as well. In conclusion,
P = L
0
RL
T
0
+ (L −L
0
)R(L −L
0
)
T
,
the sum of two positive definite or at least semidefinite matrices. For such matrices, the norm of the sum is greater or
equal to either norm, so this expression is minimized when the second term vanishes, that is, when L = L
0
.
This proves that the estimator given by (7.18) is the best, that is, that it has minimum covariance. To prove that the
covariance P of ˆx is given by equation (7.17), we simply substitute L
0
for L in P = LRL
T
:
P = L
0
RL
T
0
= (H
T
R
−1
H)
−1
H
T
R
−1
RR
−1
H(H
T
R
−1
H)
−1
= (H
T
R
−1
H)
−1
H
T
R
−1
H(H
T
R
−1
H)
−1
= (H
T
R
−1
H)
−1
as promised. ∆
94 CHAPTER 7. STOCHASTIC STATE ESTIMATION
7.5 The Kalman Filter: Derivation
We now have all the components necessary to write the equations for the Kalman filter. To summarize, given a linear
measurement equation
y = Hx + n
where n is a Gaussian random vector with zero mean and covariance matrix R,
n ∼ N(0, R) ,
the best linear unbiased estimate ˆx of x is
ˆx = PH
T
R
−1
y
where the matrix
P

= E[(ˆx −x)(ˆx −x)
T
] = (H
T
R
−1
H)
−1
is the covariance of the estimation error.
Given a dynamic system with system and measurement equations
x
k+1
= F
k
x
k
+G
k
u
k

k
(7.19)
y
k
= H
k
x
k

k
where the system noise η
k
and the measurement noise ξ
k
are Gaussian random vectors,
η
k
∼ N(0, Q
k
)
ξ
k
∼ N(0, R
k
) ,
as well as the best, linear, unbiased estimate ˆx
0
of the initial state with an error covariance matrix P
0
, the Kalman
filter computes the best, linear, unbiased estimate ˆx
k|k
at time k given the measurements y
0
, . . . , y
k
. The filter also
computes the covariance P
k|k
of the error ˆx
k|k
−x
k
given those measurements. Computation occurs according to the
phases of update and propagation illustrated in figure 7.2. We now apply the results from optimal estimation to the
problem of updating and propagating the state estimates and their error covariances.
7.5.1 Update
At time k, two pieces of data are available. One is the estimate ˆx
k|k−1
of the state x
k
given measurements up to but not
including y
k
. This estimate comes with its covariance matrix P
k|k−1
. Another way of saying this is that the estimate
ˆx
k|k−1
differs from the true state x
k
by an error term e
k
whose covariance is P
k|k−1
:
ˆx
k|k−1
= x
k
+ e
k
(7.20)
with
E[e
k
e
T
k
] = P
k|k−1
.
The other piece of data is the new measurement y
k
itself, which is related to the state x
k
by the equation
y
k
= H
k
x
k

k
(7.21)
with error covariance
E[ξ
k
ξ
T
k
] = R
k
.
We can summarize this available information by grouping equations 7.20 and 7.21 into one, and packaging the error
covariances into a single, block-diagonal matrix. Thus, we have
y = Hx
k
+ n
7.5. THE KALMAN FILTER: DERIVATION 95
where
y =
_
ˆx
k|k−1
y
k
_
, H =
_
I
H
k
_
, n =
_
e
k
ξ
k
_
,
and where n has covariance
R =
_
P
k|k−1
0
0 R
k
_
.
As we know, the solution to this classical estimation problem is
ˆx
k|k
= P
k|k
H
T
R
−1
y
P
k|k
= (H
T
R
−1
H)
−1
.
This pair of equations represents the update stage of the Kalman filter. These expressions are somewhat wasteful,
because the matrices H and R contain many zeros. For this reason, these two update equations are now rewritten in a
more efficient and more familiar form. We have
P
−1
k|k
= H
T
R
−1
H
=
_
I H
T
k
¸
_
P
−1
k|k−1
0
0 R
−1
k
_ _
I
H
k
_
= P
−1
k|k−1
+H
T
k
R
−1
k
H
k
and
ˆx
k|k
= P
k|k
H
T
R
−1
y
= P
k|k
_
P
−1
k|k−1
H
T
k
R
−1
k
_
_
ˆx
k|k−1
y
k
_
= P
k|k
(P
−1
k|k−1
ˆx
k|k−1
+H
T
k
R
−1
k
y
k
)
= P
k|k
((P
−1
k|k
−H
T
k
R
−1
k
H
k
)ˆx
k|k−1
+H
T
k
R
−1
k
y
k
)
= ˆx
k|k−1
+P
k|k
H
T
k
R
−1
k
(y
k
−H
k
ˆx
k|k−1
) .
In the last line, the difference
r
k

= y
k
−H
k
ˆx
k|k−1
is the residue between the actual measurement y
k
and its best estimate based on ˆx
k|k−1
, and the matrix
K
k

= P
k|k
H
T
k
R
−1
k
is usually referred to as the Kalman gain matrix, because it specifies the amount by which the residue must be multi-
plied (or amplified) to obtain the correction term that transforms the old estimate ˆx
k|k−1
of the state x
k
into its new
estimate ˆx
k|k
.
7.5.2 Propagation
Propagation is even simpler. Since the new state is related to the old through the system equation 7.19, and the noise
term η
k
is zero mean, unbiasedness requires
ˆx
k+1|k
= F
k
ˆx
k|k
+G
k
u
k
,
96 CHAPTER 7. STOCHASTIC STATE ESTIMATION
which is the state estimate propagation equation of the Kalman filter. The error covariance matrix is easily propagated
thanks to the linearity of the expectation operator:
P
k+1|k
= E[(ˆx
k+1|k
−x
k+1
)(ˆx
k+1|k
−x
k+1
)
T
]
= E[(F
k
(ˆx
k|k
−x
k
) −η
k
)(F
k
(ˆx
k|k
−x
k
) −η
k
)
T
]
= F
k
E[(ˆx
k|k
−x
k
)(ˆx
k|k
−x
k
)
T
]F
T
k
+E[η
k
η
T
k
]
= F
k
P
k|k
F
T
k
+Q
k
where the system noise η
k
and the previous estimation error ˆx
k|k
−x
k
were assumed to be uncorrelated.
7.5.3 Kalman Filter Equations
In summary, the Kalman filter evolves an initial estimate and an initial error covariance matrix,
ˆx
0|−1

= ˆx
0
and P
0|−1

= P
0
,
both assumed to be given, by the update equations
ˆx
k|k
= ˆx
k|k−1
+K
k
(y
k
−H
k
ˆx
k|k−1
)
P
−1
k|k
= P
−1
k|k−1
+H
T
k
R
−1
k
H
k
where the Kalman gain is defined as
K
k
= P
k|k
H
T
k
R
−1
k
and by the propagation equations
ˆx
k+1|k
= F
k
ˆx
k|k
+G
k
u
k
P
k+1|k
= F
k
P
k|k
F
T
k
+Q
k
.
7.6 Results of the Mortar Shell Experiment
In section 7.2, the dynamic system equations for a mortar shell were set up. Matlab routines available through the
class Web page implement a Kalman filter (with naive numerics) to estimate the state of that system from simulated
observations. Figure 7.3 shows the true and estimated trajectories. Notice that coincidence of the trajectories does not
imply that the state estimate is up-to-date. For this it is also necessary that any given point of the trajectory is reached
by the estimate at the same time instant. Figure 7.4 shows that the distance between estimated and true target position
does indeed converge to zero, and this occurs in time for the shell to be shot down. Figure 7.5 shows the 2-norm of the
covariance matrix over time. Notice that the covariance goes to zero only asymptotically.
7.7 Linear Systems and the Kalman Filter
In order to connect the theory of state estimation with what we have learned so far about linear systems, we now show
that estimating the initial state x
0
from the first k +1 measurements, that is, obtaining ˆx
0|k
, amounts to solving a linear
system of equations with suitable weights for its rows.
The basic recurrence equations (7.10) and (7.11) can be expanded as follows:
y
k
= H
k
x
k

k
= H
k
(F
k−1
x
k−1
+G
k−1
u
k−1

k−1
) +ξ
k
= H
k
F
k−1
x
k−1
+H
k
G
k−1
u
k−1
+H
k
η
k−1

k
= H
k
F
k−1
(F
k−2
x
k−2
+G
k−2
u
k−2

k−2
) +H
k
G
k−1
u
k−1
+H
k
η
k−1

k
7.7. LINEAR SYSTEMS AND THE KALMAN FILTER 97
5 10 15 20 25 30 35
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
true (dashed) and estimated (solid) missile trajectory
Figure 7.3: The true and estimated trajectories get closer to one another. Trajectories start on the right.
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
distance between true and estimated missile position vs. time
Figure 7.4: The estimate actually closes in towards the target.
98 CHAPTER 7. STOCHASTIC STATE ESTIMATION
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
40
norm of the state covariance matrix vs time
Figure 7.5: After an initial increase in uncertainty, the norm of the state covariance matrix converges to zero. Upwards
segments correspond to state propagation, downwards ones to state update.
= H
k
F
k−1
F
k−2
x
k−2
+H
k
(F
k−1
G
k−2
u
k−2
+G
k−1
u
k−1
) +
H
k
(F
k−1
η
k−2

k−1
) +ξ
k
.
.
.
= H
k
F
k−1
. . . F
0
x
0
+H
k
(F
k−1
. . . F
1
G
0
u
0
+. . . +G
k−1
u
k−1
) +
H
k
(F
k−1
. . . F
1
η
0
+. . . +η
k−1
) +ξ
k
or in a more compact form,
y
k
= H
k
Φ(k −1, 0)x
0
+H
k
k

j=1
Φ(k −1, j)G
j−1
u
j−1

k
(7.22)
where
Φ(l, j) =
_
F
l
. . . F
j
for l ≥ j
1 for l < j
and the term
ν
k
= H
k
k

j=1
Φ(k −1, j)η
j−1

k
is noise.
The key thing to notice about this somewhat intimidating expression is that for any k it is a linear system in x
0
, the
initial state of the system. We can write one system like the one in equation (7.22) for every value of k = 0, . . . , K,
where K is the last time instant considered, and we obtain a large system of the form
z
K
= Ψ
K
x
0
+ g
K
+ n
K
(7.23)
where
z
K
=
_
¸
_
y
0
.
.
.
y
K
_
¸
_
7.7. LINEAR SYSTEMS AND THE KALMAN FILTER 99
Ψ
K
=
_
¸
¸
¸
_
H
0
H
1
F
0
.
.
.
H
K
Φ(K −1, 0)
_
¸
¸
¸
_
g
K
=
_
¸
¸
¸
_
0
H
1
G
0
u
0
.
.
.
H
K
(Φ(K −1, 1)G
0
u
0
+. . . + Φ(K −1, K)G
K−1
u
K−1
)
_
¸
¸
¸
_
n
K
=
_
¸
_
ν
0
.
.
.
ν
K
_
¸
_ .
Without knowing anything about the statistics of the noise vector n
K
in equation (7.23), the best we can do is to
solve the system
z
K
= Ψ
K
x
0
+ g
K
in the sense of least squares, to obtain an estimate of x
0
from the measurements y
0
, . . . , y
K
:
ˆx
0|K
= Ψ

K
(z
K
−g
K
)
where Ψ

K
is the pseudoinverse of Ψ
K
. We know that if Ψ
K
has full rank, the result with the pseudoinverse is the
same as we would obtain by solving the normal equations, so that
Ψ

K
= (Ψ
T
K
Ψ
K
)
−1
Ψ
T
K
.
The least square solution to system (7.23) minimizes the residue between the left and the right-hand side under the
assumption that all equations are to be treated the same way. This is equivalent to assuming that all the noise terms in
n
K
are equally important. However, we know the covariance matrices of all these noise terms, so we ought to be able
to do better, and weigh each equation to keep these covariances into account. Intuitively, a small covariance means that
we believe in that measurement, and therefore in that equation, which should consequently be weighed more heavily
than others. The quantitative embodiment of this intuitive idea is at the core of the Kalman filter.
In summary, the Kalman filter for a linear system has been shown to be equivalent to a linear equation solver, under
the assumption that the noise that affects each of the equations has the same probability distribution, that is, that all
the noise terms in n
K
in equation 7.23 are equally important. However, the Kalman filter differs from a linear solver
in the following important respects:
1. The noise terms in n
K
in equation 7.23 are not equally important. Measurements come with covariance matrices,
and the Kalman filter makes optimal use of this information for a proper weighting of each of the scalar equations
in (7.23). Better information ought to yield more accurate results, and this is in fact the case.
2. The system (7.23) is not solved all at once. Rather, an initial solution is refined over time as new measurements
become available. The final solution can be proven to be exactly equal to solving system (7.23) all at once.
However, having better and better approximations to the solution as new data come in is much preferable in a
dynamic setting, where one cannot in general wait for all the data to be collected. In some applications, data my
never stop arriving.
3. A solution for the estimate ˆx
k|k
of the current state is given, and not only for the estimate ˆx
0|k
of the initial state.
As time goes by, knowledge of the initial state may obsolesce and become less and less useful. The Kalman
filter computes up-to-date information about the current state.

2

Chapter 1

Introduction
Fields such as robotics or computer vision are interdisciplinary subjects at the intersection of engineering and computer science. By their nature, they deal with both computers and the physical world. Although the former are in the latter, the workings of computers are best described in the black-and-white vocabulary of discrete mathematics, which is foreign to most classical models of reality, quantum physics notwithstanding. This class surveys some of the key tools of applied math to be used at the interface of continuous and discrete. It is not on robotics or computer vision, nor does it cover any other application area. Applications evolve rapidly, but their mathematical foundations remain. Even if you will not pursue any of these fields, the mathematics that you learn in this class will not go wasted. To be sure, applied mathematics is a discipline in itself and, in many universities, a separate department. Consequently, this class can be a quick tour at best. It does not replace calculus or linear algebra, which are assumed as prerequisites, nor is it a comprehensive survey of applied mathematics. What is covered is a compromise between the time available and what is useful and fun to talk about. Even if in some cases you may have to wait until you take an applied class to fully appreciate the usefulness of a particular topic, I hope that you will enjoy studying these subjects in their own right.

1.1 Who Should Take This Class
The main goal of this class is to present a collection of mathematical tools for both understanding and solving problems in fields that manipulate models of the real world, such as robotics, artificial intelligence, vision, engineering, or several aspects of the biological sciences. Several classes at most universities each cover some of the topics presented in this class, and do so in much greater detail. If you want to understand the full details of any one of the topics in the syllabus below, you should take one or more of these other classes instead. If you want to understand how these tools are implemented numerically, you should take one of the classes in the scientific computing program, which again cover these issues in much better detail. Finally, if you want to understand robotics, vision, or other applied fields, you should take classes in these subjects, since this course is not on applications. On the other hand, if you do plan to study robotics, vision, or other applied subjects in the future, and you regard yourself as a user of the mathematical techniques outlined in the syllabus below, then you may benefit from this course. Of the proofs, we will only see those that add understanding. Of the implementation aspects of algorithms that are available in, say, Matlab or LApack, we will only see the parts that we need to understand when we use the code. In brief, we will be able to cover more topics than other classes because we will be often (but not always) unconcerned with rigorous proof or implementation issues. The emphasis will be on intuition and on practicality of the various algorithms. For instance, why are singular values important, and how do they relate to eigenvalues? What are the dangers of Newton-style minimization? How does a Kalman filter work, and why do PDEs lead to sparse linear systems? In this spirit, for instance, we discuss Singular Value Decomposition and Schur decomposition both because they never fail and because they clarify the structure of an algebraic or a differential linear problem.

3

1.1 Algebraic linear systems 2.2 Weighted least squares 3.2 4.1.1 Linear estimation 3.2.1.1.2 Statistical estimation 3.3 2.3 The Kalman filter 4.1 4. but how much we cover depends on how fast we go.1 3.1 4.2 Levenberg-Marquardt method 2.1.1 Ordinary differential linear systems 3.4 3.2.3.1.1 Tensor fields of several variables 4.3 Constraints and Lagrange multipliers 3. Introduction 2. curl Line.5 Eigenvalues and eigenvectors The Schur decomposition Ordinary differential linear systems The matrix zoo Real. and volume integrals Green’s theorem and potential fields of two variables Stokes’ and divergence theorems and potential fields of three variables Diffusion and flow problems Finite differences Direct versus iterative solution methods Jacobi and Gauss-Seidel iterations Successive overrelaxation 4. positive-definite matrices 3.3 3.2 Syllabus Here is the ideal syllabus.2.3 4.4 Grad.3.2 Function optimization 2.2 4.1.1 2. symmetric.2 3.1.1.2.1. surface.5 4. 1. div.2.2 Partial differential equations and sparse linear systems 4. INTRODUCTION 1. Unknown functions of one real variable 3.1 Newton and Gauss-Newton methods 2.1.3 4.1.1.4 4.2.2 2.2 The brachistochrone .2. Unknown numbers 2.2. Unknown functions of several variables 4.4 Characterization of the solutions to a linear system Gaussian elimination The Singular Value Decomposition The pseudoinverse 2.2.2.1 Euler-Lagrange equations 4.3 Calculus of variations 4.4 CHAPTER 1.1.

1.3. DISCUSSION OF THE SYLLABUS

5

1.3 Discussion of the Syllabus
In robotics, vision, physics, and any other branch of science whose subject belongs to or interacts with the real world, mathematical models are developed that describe the relationship between different quantities. Some of these quantities are measured, or sensed, while others are inferred by calculation. For instance, in computer vision, equations tie the coordinates of points in space to the coordinates of corresponding points in different images. Image points are data, world points are unknowns to be computed. Similarly, in robotics, a robot arm is modeled by equations that describe where each link of the robot is as a function of the configuration of the link’s own joints and that of the links that support it. The desired position of the end effector, as well as the current configuration of all the joints, are the data. The unknowns are the motions to be imparted to the joints so that the end effector reaches the desired target position. Of course, what is data and what is unknown depends on the problem. For instance, the vision system mentioned above could be looking at the robot arm. Then, the robot’s end effector position could be the unknowns to be solved for by the vision system. Once vision has solved its problem, it could feed the robot’s end-effector position as data for the robot controller to use in its own motion planning problem. Sensed data are invariably noisy, because sensors have inherent limitations of accuracy, precision, resolution, and repeatability. Consequently, the systems of equations to be solved are typically overconstrained: there are more equations than unknowns, and it is hoped that the errors that affect the coefficients of one equation are partially cancelled by opposite errors in other equations. This is the basis of optimization problems: Rather than solving a minimal system exactly, an optimization problem tries to solve many equations simultaneously, each of them only approximately, but collectively as well as possible, according to some global criterion. Least squares is perhaps the most popular such criterion, and we will devote a good deal of attention to it. In summary, the problems encountered in robotics and vision, as well as other applications of mathematics, are optimization problems. A fundamental distinction between different classes of problems reflects the complexity of the unknowns. In the simplest case, unknowns are scalars. When there is more than one scalar, the unknown is a vector of numbers, typically either real or complex. Accordingly, the first part of this course will be devoted to describing systems of algebraic equations, especially linear equations, and optimization techniques for problems whose solution is a vector of reals. The main tool for understanding linear algebraic systems is the Singular Value Decomposition (SVD), which is both conceptually fundamental and practically of extreme usefulness. When the systems are nonlinear, they can be solved by various techniques of function optimization, of which we will consider the basic aspects. Since physical quantities often evolve over time, many problems arise in which the unknowns are themselves functions of time. This is our second class of problems. Again, problems can be cast as a set of equations to be solved exactly, and this leads to the theory of Ordinary Differential Equations (ODEs). Here, “ordinary” expresses the fact that the unknown functions depend on just one variable (e.g., time). The main conceptual tool for addressing ODEs is the theory of eigenvalues, and the primary computational tool is the Schur decomposition. Alternatively, problems with time varying solutions can be stated as minimization problems. When viewed globally, these minimization problems lead to the calculus of variations. When the minimization problems above are studied locally, they become state estimation problems, and the relevant theory is that of dynamic systems and Kalman filtering. The third category of problems concerns unknown functions of more than one variable. The images taken by a moving camera, for instance, are functions of time and space, and so are the unknown quantities that one can compute from the images, such as the distance of points in the world from the camera. This leads to Partial Differential equations (PDEs), or to extensions of the calculus of variations. In this class, we will see how PDEs arise, and how they can be solved numerically.

6

CHAPTER 1. INTRODUCTION

1.4 Books
The class will be based on these lecture notes, and additional notes handed out when necessary. Other useful references include the following. • R. Courant and D. Hilbert, Methods of Mathematical Physics, Volume I and II, John Wiley and Sons, 1989. • D. A. Danielson, Vectors and Tensors in Engineering and Physics, Addison-Wesley, 1992. • J. W. Demmel, Applied Numerical Linear Algebra, SIAM, 1997. • A. Gelb et al., Applied Optimal Estimation, MIT Press, 1974. • P. E. Gill, W. Murray, and M. H. Wright, Practical Optimization, Academic Press, 1993. • G. H. Golub and C. F. Van Loan, Matrix Computations, 2nd Edition, Johns Hopkins University Press, 1989, or 3rd edition, 1997. • W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes in C, 2nd Edition, Cambridge University Press, 1992. • G. Strang, Introduction to Applied Mathematics, Wellesley- Cambridge Press, 1986. • A. E. Taylor and W. R. Mann, Advanced Calculus, 3rd Edition, John Wiley and Sons, 1983. • L. N. Trefethen and D. Bau, III, Numerical Linear Algebra, SIAM, 1997. • R. Weinstock, Calculus of Variations, Dover, 1974.

Chapter 2

Algebraic Linear Systems
An algebraic linear system is a set of m equations in n unknown scalars, which appear linearly. Without loss of generality, an algebraic linear system can be written as follows: Ax = b (2.1)

where A is an m × n matrix, x is an n-dimensional vector that collects all of the unknowns, and b is a known vector of dimension m. In this chapter, we only consider the cases in which the entries of A, b, and x are real numbers. Two reasons are usually offered for the importance of linear systems. The first is apparently deep, and refers to the principle of superposition of effects. For instance, in dynamics, superposition of forces states that if force f1 produces acceleration a1 (both possibly vectors) and force f2 produces acceleration a2 , then the combined force f1 + αf2 produces acceleration a1 + αa2 . This is Newton’s second law of dynamics, although in a formulation less common than the equivalent f = ma. Because Newton’s laws are at the basis of the entire edifice of Mechanics, linearity appears to be a fundamental principle of Nature. However, like all physical laws, Newton’s second law is an abstraction, and ignores viscosity, friction, turbulence, and other nonlinear effects. Linearity, then, is perhaps more in the physicist’s mind than in reality: if nonlinear effects can be ignored, physical phenomena are linear! A more pragmatic explanation is that linear systems are the only ones we know how to solve in general. This argument, which is apparently more shallow than the previous one, is actually rather important. Here is why. Given two algebraic equations in two variables, f (x, y) = g(x, y) = we can eliminate, say, y and obtain the equivalent system F (x) = 0 y = h(x) . Thus, the original system is as hard to solve as it is to find the roots of the polynomial F in a single variable. Unfortunately, if f and g have degrees df and dg , the polynomial F has generically degree df dg . Thus, the degree of a system of equations is, roughly speaking, the product of the degrees. For instance, a system of m quadratic equations corresponds to a polynomial of degree 2m . The only case in which the exponential is harmless is when its base is 1, that is, when the system is linear. In this chapter, we first review a few basic facts about vectors in sections 2.1 through 2.4. More specifically, we develop enough language to talk about linear systems and their solutions in geometric terms. In contrast with the promise made in the introduction, these sections contain quite a few proofs. This is because a large part of the course material is based on these notions, so we want to make sure that the foundations are sound. In addition, some of the proofs lead to useful algorithms, and some others prove rather surprising facts. Then, in section 2.5, we characterize the solutions of linear algebraic systems. 7 0 0,

 b= . . We have n n xj aj = xk ak + j=1 j=1. . .3) is the same as Ax = 0 where A= a1 ··· an . . not all of which are zero. they are linearly dependent if there is a set of coefficients x1 .  x= . xn . Equation (2.  x1  . it is useful to rewrite the last two equalities in a different form. . Make sure that you are comfortable with this. . .1 Linear (In)dependence Given n vectors a1 . The converse is proven similarly: if n ak = j=1. . . j=1 Let xk be nonzero for some k. take the time to write out the components of each expression for a small example. . . . In other words.2) is the same as Ax = b and equation (2.4) If you are not convinced of these equivalences.6) as desired. xn . In one direction.” . . Proof. xn . . bm  (2. .1 The vectors a1 . . . the vector n b= j=1 x j aj (2. the columns of a matrix A are dependent if there is a nonzero solution to the homogeneous system (2. dependency means that there is a nonzero vector x such that n xj aj = 0 .2) is said to be a linear combination of a1 . j=1 (2. j=k 1 “iff” x j aj means “if and only if.1. . This is important. Vectors that are not dependent are independent.8 CHAPTER 2. The vectors a1 . . an with coefficients x1 . . . j=k xj aj xk (2. an are linearly dependent iff1 at least one of them is a linear combination of the others. . Thus. .  . . j=k n xj aj = 0 so that ak = − j=1.  . . an are linearly dependent if they admit the null vector as a nonzero linear combination.5). an and n real numbers x1 .3) For later reference. .5) (2. . . Theorem 2. xn   b1  . ALGEBRAIC LINEAR SYSTEMS 2. . . such that n xj aj = 0 .

and state the following Lemma 2. this implies that every linear space contains the zero vector. an are linearly dependent then at least one of them is a linear combination of the ones that precede it. The previous theorem is a very important result. . Let also b= j=1 n xj aj . . The basis vectors are said to span the vector space. an of A are linearly independent and the system Ax = b admits a solution.2 Basis A set a1 . . .1. 0=b−b= n n n xj aj − j=1 j=1 xj aj = j=1 (xj − xj )aj ∆ but because the aj are linearly independent. then 9 n xj aj = 0 j=1 by letting xk = −1 (so that x is nonzero). Then. . this is possible only when xj − xj = 0 for every j. Proof. which then becomes n ak = j<k xj aj xk ∆ as desired.2. B is said to be a vector space if it contains all the linear combinations of its basis vectors. . . Theorem 2. . 2. . . Just let k be the last of the nonzero xj . . an for B. In particular. the coefficients x1 . . Proof. .2. ∆2 We can make the first part of the proof above even more specific.1 Given a vector b in the vector space B and a basis a1 .2 If n nonzero vectors a1 . Then xj = 0 for j > k in (2. then the solution is unique. an is said to be a basis for a set B of vectors if the aj are linearly independent and every vector in B can be written as a linear combination of them. 2 This symbol marks the end of a proof.6). . BASIS for some k. . .2. . An equivalent formulation is the following: If the columns a1 . . xn such that n b= j=1 xj aj are uniquely determined. . . .

one of the vectors in G is a linear combination of those preceding it. Since B is a vector space. and that there are still left-over a vectors that have not been put into G. .2. . all the a s are in B.10 Pause for a minute to verify that this formulation is equivalent. since the vectors in it cannot generate the left-over a s. it makes sense to define the dimension of a space as the number of vectors in any of its bases. . . Now we can repeat the whole procedure with the roles of a vectors and a vectors exchanged. an and a1 .2 Two different bases for the same vector space B have the same number of vectors. Then each aj is in B (why?).  b= . .1. This is absurd. Suppose in fact per absurdum that G is now made only of a vectors. Since the a s form a basis. . This shows that n ≥ n. and can therefore be written as a linear combination of a1 . bm m  can be written as b= bj e j j=1 and the ej are clearly independent. . . . But then G cannot be a generating set. . The a vectors cannot run out first. ∆ A consequence of this theorem is that any basis for Rm has m vectors. the vectors of the set G = a1 . which proves that n ≥ n . So it must be one of the aj vectors. since it has no other vectors preceding it. .2. We then switch the roles of a and a vectors to conclude that n ≥ n. n ≥ n . Thus. Consequently. From lemma 2. a2 . The rest of the proof now proceeds as follows: we keep removing a vectors from G and replacing them with a vectors in such a way as to keep G a generating set for B. since any basis for Rm can only have m vectors. since any vector  b1  .  . . . an must be linearly dependent. This proves that n = n . which are independent of those in G. and the two results together imply that n = n . Now we can add a2 to G. the basis of elementary vectors ej = jth column of the m × m identity matrix is clearly a basis for Rm . Let us continue this procedure until we run out of either a vectors to remove or a vectors to add. This vector cannot be a1 . G is still a generating set for B. . Then we show that we cannot run out of a vectors before we run out of a vectors. an . Let a1 . . That is. We call a set of vectors that contains a basis for B a generating set for B. an be two different bases for B. theorem 2. writing it right after a1 : G = a1 . Proof.2. In fact. ALGEBRAIC LINEAR SYSTEMS Theorem 2.2 is that n vectors of dimension m < n are bound to be dependent. Another consequence of theorem 2. . a1 . . CHAPTER 2.2. G is a generating set for B. Removing the latter keeps G a generating set. they are mutually linearly independent. Since all bases for a space have the same number of vectors. Consequently. . . we must run out of a s first (or simultaneously with the last a). Since this elementary basis has m vectors. . . since the removed vector depends on the others.2 implies that any other basis for Rm has m vectors. because at every step we have made sure that G remains a generating set.

vectors are column vectors by default. In conclusion. c a θ b Figure 2. By Pythagoras’ theorem. Pythagoras’ theorem can now be applied again to the last sum by singling out its first term b2 e2 . projection. In the previous section we saw that any vector in Rm can be written as the linear combination m b= j=1 bj e j (2. . j This result extends Pythagoras’ theorem to m dimensions.3. How long is b? From equation (2. m b 2 = j=1 b2 . the squared length of a vector is the inner product of the vector with itself.1: The law of cosines states that a2 = b2 + c2 − 2bc cos θ. the square of the length b of b is m b 2 = b2 + 1 j=2 bj e j 2 . INNER PRODUCT AND ORTHOGONALITY 11 2. and the symbol T makes them into row vectors. because the coordinate axes are orthogonal by construction. The length of these elementary vectors is clearly one. (2.2. then b 2 = bT b . Here and elsewhere. The fundamental geometric fact that is assumed to be known is the law of cosines: given a triangle with sides a. any two of these vectors form a 90-degree angle. c (see figure 2. If we define the inner product of two m-dimensional vectors as follows: m b c= j=1 T bj cj .7) of the elementary vectors that point along the coordinate axes.3 Inner Product and Orthogonality In this section we establish the geometric meaning of the algebraic notions of norm. obtained when θ = ±π/2. A special case of this law is Pythagoras’ theorem. Also. inner product. and orthogonality.1).7) we obtain m b = b1 e1 + j=2 bj ej and the two vectors b1 e1 and m j=2 bj ej are orthogonal. we have a2 = b2 + c2 − 2bc cos θ where θ is the angle between the sides of length b and c. and so forth.8) Thus. b. because each of them goes from the origin to the unit point of one of the axes.

and b − c yields b−c 2 = b 2 + c 2 −2 b c cos θ and from equation (2. Proof.2. ∆ Given two vectors b and c applied to the origin. Since by definition point p is on the line through c.3 The projection of b onto c is the vector p = Pc b where Pc is the following square matrix: Pc = ccT . c cos θ .2 Two nonzero vectors b and c in Rm are mutually orthogonal iff bT c = 0.2: The vector from the origin to point p is the projection of b onto c. c .1 bT c = b where θ is the angle between b and c. the line between p and the endpoint of b is shortest when it is orthogonal to c: cT (b − ac) = 0 .3. the projection of b onto c is the vector from the origin to the point p on the line through c that is nearest to the endpoint of b. The line from the endpoint of b to p is orthogonal to c. CHAPTER 2. When θ = ±π/2. Proof. the previous theorem yields bT c = 0. cT c Proof. Theorem 2.12 Theorem 2. b p c Figure 2. ∆ Corollary 2.8) we obtain bT b + cT c − 2bT c = bT b + cT c − 2 b Canceling equal terms and dividing by -2 yields the desired result. where a is some real number. ALGEBRAIC LINEAR SYSTEMS c cos θ The law of cosines applied to the triangle with sides b .3.3. the projection vector p has the form p = ac. From elementary geometry. See figure 2.

. This section introduces the notion of orthogonality between spaces. It is important to understand exactly what is being mapped into what in order to determine whether a linear system has solutions. If vector space A is a subspace of Rm for some m. . .4. . while the orthogonal complement of the xy plane is the z axis. until m = n. . there is little to prove.1 Any basis a1 . an . cT b cT c ccT b cT c 13 ∆ 2. . . Proof. If r = 1. the complement of the xy plane in R3 is all of R3 except the xy plane. we also introduce a useful procedure (Gram-Schmidt) for orthonormalizing a set of linearly independent vectors. that is linearly independent of a1 .4. defines the null space and range of a matrix. .4 Orthogonal Subspaces and the Rank of a Matrix Linear transformations map spaces into spaces. This argument can be repeated until the basis spans all of Rm . If n < m.2. Theorem 2. . an for a subspace A of Rm can be extended into a basis for Rm by adding m − n vectors an+1 . an . . . . ∆ Theorem 2. With these tools. qr that span the same space as a1 . . then the orthogonal complement of A is the set of all vectors in Rm that are orthogonal to all the vectors in A. that is. . . so there must be a vector. For instance. the following construction r=0 for j = 1 to n aj = aj − if aj = 0 r =r+1 a qr = aj j end end r T l=1 (ql aj )ql yields a set of orthonormal 3 vectors q1 .2 (Gram-Schmidt) Given n vectors a1 . If n = m we are done. The normalization in the above procedure ensures that q1 has unit norm. an . . . . Notice that complement and orthogonal complement are very different notions. . .5. Proof. . and if so how many. Let us now assume that the procedure 3 Orthonormal means orthogonal and with unit norm. the given basis cannot generate all of Rm . We first prove by induction on r that the vectors qr are mutually orthonormal. am . . Two vector spaces A and B are said to be orthogonal to one another when every vector in A is orthogonal to every vector in B. call it an+1 . ORTHOGONAL SUBSPACES AND THE RANK OF A MATRIX which yields a= so that p = ac = c a = as advertised. In the process.4. we will be able to characterize the solutions to a linear system in section 2. and its rank. . .

. . . the dimension of the orthogonal space A⊥ is exactly m − n. . are orthonormal (and therefore independent). must be some linear combination of v1 . . Avn are a basis for the range of A. .4. qn span A. and b ∈ range(A) iff Ax = b for some x. . Because the new basis is orthonormal. . being an element of Rn . In fact. there is an n-tuple x such that Ax = b. . . .1) into a basis v1 . . Then for any i < r we have r−1 qT aj = qT aj − i i l=1 (qT aj )qT ql = 0 l i because the term cancels the i-th term of the sum (remember that qT qi = 1). an be a basis for A. vh be a basis for null(A). and extend this basis (theorem 2. vn for Rn . First. . . the vector qr . ∆ qT aj i (qT aj )qT qi i i Theorem 2. Proof. . . if null(A) has dimension h. . On the other hand. . and because qT aj = 0. . Finally. . qT qr = 0 for i i i = 1. am for Rm (theorem 2. . . Because of the explicit normalization step qr = aj / aj . and equal in number to the number of linearly independent vectors in a1 . Let v1 .1). .3 If A is a subspace of Rm and A⊥ is the orthogonal complement of A in Rm . Extend this basis to a basis a1 . Now we show that the number of independent columns is also n − h. . . because otherwise we would have more than m vectors in a basis for Rm . our basis for Rn : n x= j=1 cj vj . all vectors generated by qn+1 . . if computed. . .4. . q1 . . The n-tuple x itself. A has n − h linearly independent rows. then dim(A) + dim(A⊥ ) = m . . . This is the point of the following theorem. and the inner products i T qi ql are zero by the inductive assumption. r − 1. and that these vectors are orthonormal (the inductive assumption). Then we can show that the n − h vectors Avh+1 . The range of A is the space of all m-dimensional vectors that are generated by the columns of A.4. Thus. and r =n−h where h = dim(null(A)). . the dimension of this orthogonal space cannot exceed m − n. qn . . From theorem 2. there must be a linear combination of the columns of A that is equal to b. Proof. . then the space generated by the rows of A has dimension r = n − h. by constructing a basis for range(A). The null space null(A) of an m × n matrix A is the space of all n-dimensional vectors that are orthogonal to the rows of A. we notice that the vectors qj span the same space as the aj s. Thus. . . . . given an arbitrary vector b ∈ range(A). . ALGEBRAIC LINEAR SYSTEMS above has been performed a number j − 1 of times sufficient to find r − 1 vectors q1 . . We have already proven that the number of independent rows is n − h.14 CHAPTER 2. x ∈ null(A) iff Ax = 0. By construction. qm . . because the former are linear combinations of the latter. . .4. vn .4. it follwos that qr is orthogonal to all its predecessors. . . as promised. has unit norm. Orthonormalize this basis by the Gram-Schmidt procedure (theorem 2. .2) to obtain q1 .4. . In symbols. . that is. Let a1 . qm are orthogonal to all vectors generated by q1 .4 The number r of linearly independent columns of any m × n matrix A is equal to the number of its independent rows. ∆ We can now start to talk about matrices in terms of the subspaces associated with them. . . It is not obvious that the space generated by the columns of A has also dimension r = n − h. so there is a space of dimension at least m − n that is orthogonal to A. .3. an . . qr−1 . these n − h vectors generate the range of A. Theorem 2.

2. dimension h range(A)⊥ . . not all zero. . . . such that n xj Avj = 0 j=h+1 so that A n xj vj = 0 . . Second. . This proves that the n − h vectors Avh+1 . defined as the matrix obtained by exchanging the rows of A with its columns. . there must exist coefficients x1 . . . . xh such that n h n xj vj = j=h+1 j=1 xj vj . vn are linearly independent. dimension r = n − h . . Theorem 2. and null(A)⊥ is called the rowspace of A. Then there exist numbers xh+1 . . It should be obvious from the meaning of these spaces that null(A)⊥ range(A)⊥ = range(AT ) = null(AT ) where AT is the transpose of A. h. . . . The space range(A)⊥ is called the left nullspace of the matrix. . . . Avn are linearly independent. . we can define the rank of A to be equivalently the number of linearly independent columns or of linearly independent rows of A: rank(A) = dim(range(A)) = n − dim(null(A)) . Suppose. vh span null(A). j=h+1 But then the vector j=h+1 xj vj is in the null space of A. Thanks to this theorem. so that Avj = 0 for j = 1. Avn generate range(A). xn . b = Ax = A j=1 15 n n n cj vj = j=1 cj Avj = j=h+1 cj Avj since v1 . A frequently used synonym for “range” is column space. vh are a basis for null(A). dimension r = rank(A) null(A). we prove that the n − h vectors Avh+1 . dimension m − r null(A)⊥ . . per absurdum. . . .5 The Solutions of a Linear System Thanks to the results of the previous sections. . ∆ in conflict with the assumption that the vectors v1 . .5.5. 2. and an arbitrary vector x into a vector in range(A).1 The matrix A transforms a vector x in its null space into the zero vector. . THE SOLUTIONS OF A LINEAR SYSTEM Thus. . . . we now have a complete picture of the four spaces associated with an m × n matrix A of rank r and null-space dimension h: range(A). . . . . that they are not. Since the vectors v1 .

and particularly its version called reduction to echelon form is such a method. Thus. All the cases are summarized in figure 2. but since b is in the range of A there is a linear combination of the columns (a vector x) that produces b. In the first case above. let r = rank(A) be the number of linearly independent rows or columns of A. so the first two cases exhaust the possibilities for r = n. the equations are compatible. Notice that invertibility depends only on A. Gaussian elimination. no matter whether the system is invertible or not.. and on numerical considerations.16 CHAPTER 2.3. like Gaussian elimination. 2.e. In the second. three possibilities occur. This means that there is exactly one x that satisfies the system. Then. and exactly one solution exists. Let Ax = b be an m × n system (m can be less than. and is summarized in the next section. Of course.. b. are done in a finite amount of time that can be bounded given only the size of a matrix. Also. n: • When r = n = m. Notice that if r = n then n cannot possibly exceed m. There are more equations than unknowns. while direct methods.g. ALGEBRAIC LINEAR SYSTEMS This allows characterizing the set of solutions to linear system as follows. “redundant” means “with exactly one solution and with more equations than unknowns. In addition to always yielding a solution.” . Which method to use depends on the size and structure (e.9) 4 Notice that the technical meaning of “redundant” has a stronger meaning than “with more equations than unknowns. there can be no linear combination of the columns (no x vector) that gives b. Also. b ∈ range(A) b ∈ range(A) ⇒ no solutions ⇒ ∞n−r solutions with the convention that ∞0 = 1. ∞k is the cardinality of a k-dimensional vector space. listing all possibilities does not provide an operational method for determining the type of linear system for a given pair A. • When r = n and m > n. but then for any y ∈ null(A) also x + y is a solution: Ax = b . sparsity) of the matrix. depending on the relative sizes of r. and there is a space of dimension h = n − r of vectors x such that Ax = 0. or greater than n). In other words. Consider the m × n system Ax = b (2. r cannot exceed either m or n. Other solution techniques exist for linear systems. the system is redundant. it also allows determining the rank of a matrix. it has dimension h > 0). Here. the system is invertible. Since b is assumed to be in the range of A.” The case r < n < m is possible. not on b. since the columns of A span all of Rn . iterative methods solve systems in a time that depends on the accuracy required. 4 • When r < n the system is underdetermined.6 Gaussian Elimination Gaussian elimination is an important technique for solving linear systems. and the system is said to be incompatible. there are solutions x to Ax = b. but is called “underdetermined” because there are fewer (r) independent equations than there are unknowns (see next item). admits a solution if b ∈ range(A). compatible case. Most notably. equal to. Ay = 0 ⇒ A(x + y) = b and this generates the ∞h = ∞n−r solutions mentioned above. This means that the null space is nontrivial (i. m. whether more information is required about the matrix of the system. More on this in chapter 3. has more equations (m) than unknowns (n).

3: Types of linear systems. incompatible. or underdetermined. Initially. ways of selecting l here lead to different numerical properties of the algorithm. which is transformed to a new vector c. In particular. U is in echelon form. Once the system is transformed into echelon form. This zeros all the entries in the column below the pivot. Gaussian elimination replaces the rows of this system by linear combinations of the rows themselves until A is changed into a matrix U that is in the so-called echelon form. The first nonzero entry. Here is step k. 6 Different 5 “Stop” . m. of a row. For i = k + 1. . . j of U (k) : Skip no-pivot columns If uip is zero for every i = k. • Below each pivot is a column of zeros. . If l = k.5 Row exchange Now p ≤ n and uip is nonzero for some k ≤ i ≤ m. Let l be one such value of i6 . The same operations are applied to the rows of A and to those of b. so equality is preserved and solving the final system yields the same solution as solving the original one. that is. If p exceeds n stop. if the matrix is square and if all columns have a pivot. the “pivot column index” p is set to one. . m of U (k) and c(k) and produces U (k+1) and c(k+1) . by solving the transformed system Ux = c . then increment p by 1. where uij denotes entry i. When this process is finished. if any. exchange rows l and k of U (k) and of c(k) . In short. and is called the pivot. and preserves the equality of left. . . we compute the solution x by backsubstitution. . The k-th step is applied to rows k.2. there are no restrictions on the system. Selecting the largest entry in the column leads to better round-off properties. . . and subtract entry k of c(k) multiplied by uip /ukp from entry i of c(k) . subtract row k of U (k) multiplied by uip /ukp from row i of U (k) . . redundant. then U is upper-triangular. This means that • Nonzero rows precede rows with all zeros. The last step produces U (m) = U and c(m) = c. The first step is applied to U (1) = A and c(1) = b. 2. m.and right-hand side. means that the entire algorithm is finished.1 Reduction to Echelon Form The matrix A is reduced to echelon form by a process in m − 1 steps.6. is called a pivot. GAUSSIAN ELIMINATION b in range(A) yes r=n yes m=n yes invertible no redundant Figure 2. . • Each pivot lies to the right of the pivot in the row above. 17 no incompatible no underdetermined which can be square or rectangular. . invertible. Triangularization The new entry ukp is nonzero.6.

however. 0 cm Let us call this the residual subsystem. because A and U admit exactly the same solutions and are equal in size. .   . The final result is a solution with as many free parameters as there are free variables. . that is.10) satisfied.11). . Since no vector x can satisfy these equations. . We are left with an r × n system. Then. Since this is the number of independent rows of U . r is the rank of U . the last m − r equations yield a subsystem of the following form:     cr+1 0  . To see this. . jr . and therefore of A. Since any value given to the free variables leaves the equality of system (2.e. and call the other n − r the free variables. If on the other hand r = m (obviously r cannot exceed m). If r < m.. ALGEBRAIC LINEAR SYSTEMS 2. . or if there is one it is compatible. . . .11) where the xji are the free variables. cm are nonzero. It is also the rank of A. Similarly. remove the residual system. . the vector vi for i = 1. 7 An n l=ji +1  uil xl  affine function is a linear function plus a constant.2 Backsubstitution A system Ux = c (2. and then transform this solution procedure into one that can be more readily implemented numerically. n − r can be obtained by solving the homogeneous system Ux = 0 with xji = 1 and all other free variables equal to zero. once for the nonhomogeneous system. When solving a system in echelon form numerically. In this system. . . First. Whenever they appear in the expressions for the basic variables. Say that the pivot columns are j1 . The vector v0 is the solution when all free variables are zero. the presence of free variables leads to an infinity of solutions. the general solution can be obtained by running backsubstitution n − r + 1 times. and n − r times for the homogeneous system. we first solve the system symbolically. . solutions exist. The solution obtained by backsubstitution is an affine function7 of the free variables. cr+1 = . call the variables corresponding to the r columns with pivots the basic variables. and can therefore be obtained by replacing each free variable by zero during backsubstitution. Backsubstitutions works as follows.  . . Let r be the index of the last nonzero row of U . = . . . vn−r form a basis for the null space of U . . free variables are specified by name rather than by value. it is inconvenient to carry around nonnumeric symbol names (the free variables). Then symbolic backsubstitution consists of the following sequence: for i = r downto 1  1  xji = ci − uiji end This is called symbolic backsubstitution because no numerical values are assigned to free variables. = cm = 0. . This yields the solution in the form (2. because they are of the form 0 = ci with ci = 0. .18 CHAPTER 2. . leaving undetermined variables specified by their name. . by solving the equations starting from the last one and replacing the result in the equations higher up. Here is an equivalent solution procedure that makes this unnecessary. with suitable values of the free variables. r < m) and some of cr+1 . + xjn−r vn−r (2. . there is no residual subsystem. If there is a residual system (i. In conclusion. if any. Let us now assume that either there is no residual system.6. the linear system admits no solutions: it is incompatible. Notice that the vectors v1 . and can therefore be written in the form x = v0 + xj1 v1 + .   .10) in echelon form is easily solved for x. that is. and they can be determined by backsubstitution. then the equations corresponding to these nonzero entries are incompatible. .

When applied to both U (1) and c(1) this yields     1 3 3 2 1 U (2) =  0 0 3 1  . the number of nonzero rows. c = c(3) =  3  . with ∞n−r = ∞2 solutions. row 2 multiplied by 6/3 is subtracted from row 3 for both U (2) and c(2) to yield     1 3 3 2 1 U = U (3) =  0 0 3 1  . In the triangularization step.3 An Example An example will clarify both the reduction to echelon form and backsubstitution. c(2) =  3  . GAUSSIAN ELIMINATION 19 2. In 8 other words. for both p = 1 and p = 2. the residual subsystem is first deleted. and the rank of U and that of A is r = 2. The triangularization step subtracts row 1 multiplied by 2/1 from row 2. Backsubstitution applied first to row 2 and then to row 1 yields the following expressions for the pivot variables: x3 x1 1 1 1 (c2 − u24 x4 ) = (3 − x4 ) = 1 − x4 u23 3 3 1 1 = (c1 − u12 x2 − u13 x3 − u14 x4 ) = (1 − 3x2 − 3x3 − 2x4 ) u11 1 = 1 − 3x2 − (3 − x4 ) − 2x4 = −2 − 3x2 − x4 =     −3      + x2  1  + x4    0   0   −1 0   . are free. x2 and x4 . 0 0 0 0 0 There is one zero row in the left-hand side. so the pivot column index p stays at 1. Consider the system Ax = b where U (1)   3 2 9 5  . so no row exchange is necessary.12) (2) (2) The basic variables are x1 and x3 . For k = 1. . The residual system is 0 = 0 (compatible).6. In symbolic backsubstitution. this means that u11 = a11 = 1 is the pivot. we choose a trivial pivot selection rule: we pick the (1) first nonzero entry at or below row k in the pivot column. and r < n = 4. This yields the reduced system 1 3 0 0 3 3 2 1 x= 1 3 (2. and subtracts row 1 multiplied by -1/1 from row 3. so p is set to 3: the second pivot column is column 3. 0 0 6 2 6 Notice that now (k = 2) the entries uip are zero for i = 2. 5  1 =A= 2 −1 3 6 −3 c(1) Reduction to echelon form transforms A and b as follows. so the system is underdetermined. −1  3 1 so the general solution is   −2 − 3x2 − x4 −2    0 x2 = x= 1    1 1 − 3 x4 0 x4 8 Selecting the largest entry in the column at or below row k is a frequent choice. no row exchange is necessary. and this would have caused rows 1 and 2 to be switched.6. 3. 3 0  1 =b= 5  .2. corresponding to the columns with pivots. Throughout this example. In the first step (k = 1). there are no no-pivot columns. The other two variables. and u23 is nonzero.

As mentioned at the beginning of this section. in the sense that the answer can be found in a number of steps that depends only on the size of the matrix A. 0  Then v1 is found by solving the nonzero part (first two rows) of U x = 0 with x2 = 1 and x4 = 0 to obtain x3 x1 so that = = 1 (−1 · 0) = 0 3 1 (−3 · 1 − 3 · 0 − 2 · 0) = −3 1  −3  1   v1 =   0  .20 CHAPTER 2. Solving the reduced system (2. 0  Finally. we study a different . In the next chapter. ALGEBRAIC LINEAR SYSTEMS This same solution can be found by the numerical backsubstitution method as follows. solving the nonzero part of U x = 0 with x2 = 0 and x4 = 1 leads to x3 x1 so that = = 1 1 (−1 · 1) = − 3 3 1 1 (−3 · 0 − 3 · − 1 3  −1  0  v2 =  1   −  3 1   −2  0     x = v0 + x2 v1 + x4 v2 =   1  + x2  0    −3  1   + x4   0  0  −1 0   −1  3 1  − 2 · 1) = −1 and just as before.12) with x2 = x4 = 0 by numerical backsubstitution yields x3 x1 so that = = 1 (3 − 1 · 0) = 1 3 1 (1 − 3 · 0 − 3 · 1 − 2 · 0) = −2 1  −2  0   v0 =   1  . Gaussian elimination is a direct method.

meaning that an exact solution usually requires an infinite number of steps.6. one can compute several thousand SVDs of 5 × 5 matrices in one second. This state of affairs would seem to favor Gaussian elimination over the SVD. .2. b are small numbers that depend on the details of the algorithm. very efficient algorithms for the SVD exist. the computation of the SVD is numerically well behaved. This is an iterative method. Finally. In addition. based on the so-called the Singular Value Decomposition (SVD). For instance. More generally. on a regular workstation. Singular values also allow defining a notion of approximate rank which is very useful in a large number of applications. which provide great insight into the behavior of the linear transformation represented by the matrix A. much more so than Gaussian elimination. called the singular values. However. GAUSSIAN ELIMINATION 21 method. the latter yields a much more complete answer. It also allows finding approximate solutions when the linear system in question is incompatible. and the number of steps necessary to find an approximate solution depends on the desired number of correct digits. as well as a set of quantities. since it computes bases for all the four spaces mentioned above. the number of floating point operations necessary to compute the SVD of an m × n matrix is amn2 + bn3 where a. for reasons that will become apparent in the next chapter.

ALGEBRAIC LINEAR SYSTEMS .22 CHAPTER 2.

1. vn be an orthonormal basis for S. This describes the action that a matrix has on the magnitudes of vectors as well.Chapter 3 The Singular Value Decomposition In section 2. This is important. it must be possible to write p as a linear combination of the vj s. To this end. In other words. pm and since P is in S. into the orthogonal complement of the range. . . 23 . . we first need to introduce the notion of orthogonal matrices. and let v1 . there must exist coefficients   q1  . 3. we make this statement more specific by showing how unit vectors1 in the rowspace are transformed by matrices. in section 3. i since the vj are orthonormal.1 Orthogonal Matrices Let S be an n-dimensional subspace of Rm (so that we necessarily have n ≤ m). .  p= . and vectors in its null space into the zero vector. we saw that a matrix transforms vectors in its domain into vectors in its range (column space). The chapter concludes with some basic applications and examples. .  . and interpret them geometrically as transformations between systems of orthonormal coordinates. Then for any i = 1. If the coordinates of P in Rm are collected in an m-dimensional vector   p1  .  . n we have n n vT p = vT i i j=1 qj vj = j=1 q j vT vj = q i . . . Consider a point P in S. In this section. + qn vn = V q where V = v1 ··· vn is an m × n matrix that collects the basis for S as its columns. qn such that p = q1 v1 + . that is. No nonzero vector is mapped into the left null space. . Then. and may need emphasis: 1 Vectors with unit norm. .  q= . .2. . we use these new concepts to introduce the all-important concept of the Singular Value Decomposition (SVD). We do this in section 3.

3) still holds. We can summarize this discussion as follows: Theorem 3. ⇒ V −1 V = V T V = V V −1 = V V T = I . However. j In matrix form. . so V R can have at most n linearly independent columns. the distinction between left and right inverse vanishes. Since the left inverse of a matrix V is defined as the matrix L such that LV = I . In particular. this result is still valid when V is m × m and has orthonormal columns.3) is said to be orthogonal. a matrix is orthogonal if its columns are orthonormal. and V T is then simply said to be the inverse of V : V T = V −1 .4) is unique.1 The left inverse of an orthogonal m × n matrix V with m ≥ n exists and is equal to the transpose of V: V TV = I . A matrix V that satisfies equation (3. vn are orthonormal. (3.3) shows that the left inverse of an orthogonal matrix V exists. . full-rank matrices (r = m = n). In fact. the matrix V −1 = V T is also the right inverse of V : V square 2 Nay. because it is made of orthonormal columns. the argument above shows that the rows of a square orthogonal matrix are orthonormal as well. Since the matrix V V T contains the inner products between the rows of V (just as V T V is formed by the inner products of its columns).24 If CHAPTER 3. Thus. Notice that V R = I cannot possibly have a solution when m > n. Of course.4) comparison with equation (3.1. for square. THE SINGULAR VALUE DECOMPOSITION n p= j=1 qj vj and the vectors of the basis v1 . Then L = L(V R) = (LV )R = R.2) V TV = I where I is the n × n identity matrix. while the columns of V R are linear combinations of the n columns of V . . this argument requires V to be full rank. so that the solution L to equation (3. if m = n. because the m × m identity matrix has m linearly independent 2 columns. Also. so the left and the right inverse are the same. since equation (3. . V T is both the left and the right inverse: V TV = V V T = I .3) 2 (3. However. Thus. we can collect the n equations vT vj = i into the following matrix equation: 1 0 if i = j otherwise (3.1) q = V Tp . orthonormal. then the coefficients qj are the signed magnitudes of the projections of p onto the basis vectors: qj = vT p . (3. for square orthogonal matrices. V is certainly full rank. suppose that there exist matrices L and R such that LV = I and V R = I. and is equal to the transpose of V . Of course. .

Theorem 3.1. the result is the same. . of point P with coordinates p into a different point Q with coordinates q.3 Consistently with either of these geometric interpretations. because two interpretations of it are possible. This notion of projection can be extended from lines to vector spaces by the following definition: The projection p of a point b ∈ Rn onto a subspace C is the point in C that is closest to b. Also. ∆ We conclude this section with an obvious but useful consequence of orthogonality. Then the matrix U U T projects any vector b onto range(U ). or if you stand still and the whole universe spins counterclockwise around you. U T U is the identity matrix. and the underlying reference frame is changed from the elementary vectors ej (that is. Vx 2 = xT V T V x = xT x = x 2 .3 we defined the projection p of a vector b onto another vector c as the point on the line through c that is closest to b. One solution may be more efficient than the other in other ways. ORTHOGONAL MATRICES 25 Sometimes. The derivative of this squared distance with respect to x is the vector 2x − 2U T b 3 At least geometrically. Proof. In the interpretation given above. equation (3. when m = n. and should not be surprising: If you spin clockwise on your feet.2) can be seen as a transformation.2) causes confusion. we have the following result: Theorem 3.3). is relativity. In section 2. so b−p 2 = bT b + xT x − 2bT U x .2 The norm of a vector x is not changed by multiplication by an orthogonal matrix V : Vx = x . in a fixed reference system.1. to the columns of V ). An analogous result holds for subspace projection. from the columns of I) to the vectors vj (that is. the point P remains the same. A point p in range(U ) is a linear combination of the columns of U : p = Ux where x is the vector of coefficients (as many coefficients as there are columns in U ). and the vector b − p is orthogonal to c. as the following theorem shows.3.3. the projection matrix is ccT (theorem 2.3 Let U be an orthogonal matrix. Alternatively. Proof. the geometric interpretation of equation (3. This. Because of orthogonality. however. Furthermore. The squared distance between b and p is b − p 2 = (b − p)T (b − p) = bT b + pT p − 2bT p = bT b + xT U T U x − 2bT U x . for unit vectors c. the difference vector between b and its projection p onto range(U ) is orthogonal to range(U ): U T (b − p) = 0 .1.

in one dimension it is a pair of points. the hyperellipse in question is centered at the origin. The geometric picture underlying the Singular Value Decomposition is crisp and useful. which is a quadratic hypersurface that generalizes the two-dimensional notion of ellipse to an arbitrary number of dimensions. or when geometric intuition provides a strong guiding element. This approach is successful when geometry is less cumbersome than algebra. In all cases. ∆ 3. THE SINGULAR VALUE DECOMPOSITION b3 x v2 x 1 v1 u3 σ u2 2 σ u1 1 b b 1 b 2 Figure 3.5) maps a circle on the plane into an ellipse in space. when p = Ux = UUT b as promised. For this value of p the difference vector b − p is orthogonal to range(U ).2 The Singular Value Decomposition In these notes. For instance. in the sense that U T (b − p) = U T (b − U U T b) = U T b − U T b = 0 . In three dimensions. because it also describes what happens to the magnitudes of the vectors: a hypersphere is stretched or compressed into a hyperellipse. Here is the main intuition: An m × n matrix A of rank r maps the r-dimensional unit hypersphere in rowspace(A) into an rdimensional hyperellipse in range(A). the rank-2 matrix  √ √  3 1  3 A= √ (3. The two small boxes are corresponding points. Figure 3. which is zero iff x = UT b . .5) −3 3  2 1 1 transforms the unit circle on the plane into an ellipse embedded in three-dimensional space. so we will use geometric intuition again. This statement is stronger than saying that A maps rowspace(A) into range(A). the hyperellipse is an ellipsoid.1: The matrix in equation (3. we have often used geometric intuition to introduce new concepts. and we have then translated these into algebraic statements.1 shows the map b = Ax .26 x 2 CHAPTER 3. that is.

so the first entry of U1 AV1 is uT σ1 u1 = σ1 . If this were not the case. its proof by geometric means is cumbersome. the first column of AV1 is Av1 = σ1 u1 . Theorem 3. Let x and y be unit vectors in Rn and Rm . possibly at more than one point 4 . . . . by construction. . . . .2. and let σ1 be the corresponding value of z: max yT Ax = uT Av1 = σ1 .2. we will prove it algebraically by first introducing the existence of the SVD and then using the latter to prove that matrices map hyperspheres into hyperellipses. and 1 1 therefore orthogonal to v2 . we see that v1 is parallel to AT u1 . . A similar argument shows that the entries after the first in the first row of S1 are zero: the row vector uT A is parallel to vT . 1 x = y =1 It is easy to see that u1 is parallel to the vector Av1 . We can repeat the same construction on A1 and write T U2 A1 V2 = S2 = 4 Actually. 1 1 The matrix A1 has one fewer row and column than A. . and consider the bilinear form z = yT Ax . . Simple and fundamental as this geometric fact may be.2. by noticing that uT Av1 = vT AT u1 1 1 and repeating the argument above. respectively. 1 . . thereby contradicting the fact that uT Av1 is a maximum. This result can be generalized to any m × n matrix. where p = min(m. so that the scalar function z(x. .4. By theorems 2. Then T U1 AV1 = S1 = σ1 0 0T A1 . and its other entries 1 T are uj Av1 = 0 because Av1 is parallel to u1 and therefore orthogonal. so that uT Av2 = . y | x ∈ Rn . T In fact. Collect these orthonormal basis vectors into orthogonal matrices U1 and V1 . σp ) ∈ Rm×n A = U ΣV T .1 If A is a real m × n matrix then there exist orthogonal matrices U V such that = = u1 v1 ··· ··· um vn ∈ Rm×m ∈ Rn×n U T AV = Σ = diag(σ1 . to u2 . THE SINGULAR VALUE DECOMPOSITION 27 Two diametrically opposite points on the unit circle are mapped into the two endpoints of the major axis of the ellipse. . . u1 and v1 can be extended into orthonormal bases for Rm and Rn . y) must achieve a maximum value on S. The lines through these two pairs of points on the unit circle are always orthogonal. so is (−u1 )T A(−v1 ). respectively. vn . . their inner product uT Av1 could 1 be increased by rotating u1 towards the direction of Av1 . n) and σ1 ≥ . Instead.3.4. Equivalently.1 and 2. 1 Similarly. ≥ σp ≥ 0. = uT Avn = 0. and two other diametrically opposite points on the unit circle are mapped into the two endpoints of the minor axis of the ellipse. Let u1 . v1 be two unit vectors in Rm and Rn respectively where this maximum is achieved. . Proof. The set S = {x. y ∈ Rm . σ2 0 0T A2 at least at two points: if uT Av1 is a maximum. x = y = 1} is compact. um .

. The σi are nonnegative because all these maximizations are performed on unit hyper-spheres. and is therefore orthogonal to v1 . A2 1 0 0 V2 T σ1 = 0 0 0 σ2 0 This procedure can be repeated until Ak vanishes (zero rows or zero columns) to obtain U T AV = Σ where U T and V are orthogonal matrices obtained by multiplying together all the orthogonal matrices used in the procedure. but order v1 .1. The corresponding axis unit vectors u1 . ˆ The vector x is equal to the unit vector [0 x]T transformed by the orthogonal matrix V1 . .28 so that 1 0 0 T U2 T T U1 AV1 CHAPTER 3. Since matrices U and V are orthogonal. To show this. A similar argument shows that y is a unit vector orthogonal to u1 . Because x and y thus defined belong to subsets (actually sub-spheres) of the unit spheres in Rn and Rm . and Σ = diag(σ1 . . because axis endpoints come in pairs. v2 the two right singular vectors of A. . in this order. . . This transformation can be achieved in three steps (see figure 3. It only remains to show that the elements on the diagonal of Σ are nonnegative and arranged in nonincreasing order. THE SINGULAR VALUE DECOMPOSITION   0T 0T  . we introduce some nomenclature for the three matrices in the SVD. The σi s are maxima of the function z(x. and imagine transforming point x (the small box at x on the unit circle) into its corresponding point b = Ax (the small box on the ellipse). . . . so is −x. Let us call v1 . u2 on the ellipse are called left singular vectors. To see that σ1 ≥ . There are a few ways to do this. we conclude that σ2 ≤ σ1 . n)). x = y =1 xT v1 = yT u1 = 0 To explain the last equality above. which is the desired result. vn . Write x in the frame of reference of the two vectors v1 . v2 on the unit circle that map into the major axes of the ellipse. y) which always assumes both positive and negative values on any hyper-sphere: If z(x. we can observe that the successive maximization problems that yield σ1 . σp ) . . . then z(−x. y) is negative.1 in light of the singular value decomposition. . and induction will do the rest. represented by equation (3. Just pick one way. . . v2 so they map into the major and the minor axis. and is therefore itself a unit vector. Consider the map in figure 3.5). We have σ2 = = ˆ x max ˆT A1 ˆ = max y x ˆ ˆ ˆ = y =1 x = y =1 max 0 ˆ y T T U1 AV1 0 0 ˆ x ˆ y = T S1 0 ˆ x ˆ ˆ x = y =1 max yT Ax ≤ σ1 . σp are performed on a sequence of sets each of which contains the next. we just need to show that σ2 ≤ σ1 . ≥ σp (where p = min(m. ∆ We can now review the geometric picture in figure 3. it is a linear combination of v2 . we can premultiply the matrix product in the theorem by U and postmultiply it by V T to obtain A = U ΣV T . y) is positive. In addition.2): 1. consider the vectors x = V1 0 ˆ x and y = U1 0 ˆ y . and if x is on a hyper-sphere. If we define V = v1 v2 . In the process. .

the “straight” ellipse has the same shape as the ellipse in figure 3. Rotate the reference frame in Rm = R3 so that the “straight” ellipse becomes the ellipse in figure 3. y2 axes.2: Decomposition of the mapping in figure 3. u1 u2 u3 . u3 that identify the axes of the ellipse and the normal to the plane of the ellipse. The real. σ2 are called the singular values of A. THE SINGULAR VALUE DECOMPOSITION x 2 29 y3 x v1 v2 x 1 u3 σu 2 2 σ u1 1 y y 1 y η 2 ξ 2 2 v’2 ξ σ u’ 2 2 v’1 ξ1 η σ u’ 1 1 η1 Figure 3.1. “Straight” here means that the axes of the ellipse are aligned with the y1 .1. Transform ξ into its image on a “straight” version of the final ellipse. the new coordinates ξ of x become ξ = V Tx because V is orthogonal. and maps it to b. If the lengths of the half-axes of the ellipse are σ1 . 3. σ2 (major axis first). so b = Uη where the orthogonal matrix U= collects the left singular vectors of A. u2 . the transformed vector η has coordinates η = Σξ where σ1 Σ= 0 0   0 σ2  0 is a diagonal matrix. The components of η are the signed magnitudes of the projections of b along the unit vectors u1 . nonnegative numbers σ1 . Otherwise.3.2. 2.1. This rotation brings η along.

. For instance. However. Thus. provided that the corresponding left singular vector is flipped as well. Except for these ambiguities. it can be removed by imposing. The second source of ambiguity is deeper. The first is in the orientation of the singular vectors. Similarly. . . . . if m < n. 1 : p). the subspaces identified by the corresponding left and right singular vectors are unique. minimal. so that the last m − n columns of U are multiplied by zero. then rank(A) = r null(A) = span{vr+1 . This suggests a “small. that is. This is the SVD of A. but any orthonormal basis can be chosen within.” equivalent version of the SVD. = 0 . the bottom (m − n) × n block of Σ is zero. One can flip any right singular vector. while U and V are square. i which is an even smaller. and Vr = V (:. . if m > n. if σr is the smallest nonzero singular value of A. . the axes of the latter are not defined. vn } range(A) = span{u1 . If we define r by σ1 ≥ . 1 : r). Σ is m × n. Σ has the same shape and size as A. Σp is p × p. then we have r A = Ur Σr VrT = i=1 σi ui vT . Σp = Σ(1 : p. More generally. all of the form I = U IU T where U is any orthogonal matrix of suitable size. Finally. and still obtain a valid SVD. This is a trivial ambiguity. and Vp = V (:. the identity matrix has an infinity of SVDs. There are two sources of ambiguity. . . The singular value decomposition is “almost unique”. The SVD reveals a great deal about the structure of a matrix. ≥ σr > σr+1 = . we can let Ur = U (:. the rightmost m × (n − m) block of Σ is zero. THE SINGULAR VALUE DECOMPOSITION We can concatenate these three transformations to obtain b = U ΣV T x or A = U ΣV T since this construction works for any point x on the unit circle. and Vp is n × p. If the matrix A maps a hypersphere into another hypersphere. and this multiplies the last n − m rows of V . that the first nonzero entry of every left singular value be positive. say. a valid SVD. 1 : r). Singular vectors must be flipped in pairs (a left vector and its corresponding right vector) because the singular values are required to be nonnegative. Moreover. the right subspace and yield. 1 : r). SVD. The sizes of the matrices in the SVD are as follows: U is m × m. we can define Up = U (:. Σr = Σ(1 : r. n). the singular values of a matrix A are the lengths of the semi-axes of the hyperellipse E defined by E = {Ax : x = 1} . for instance.30 CHAPTER 3. 1 : p). . 1 : p). If p = min(m. the SVD is unique. Even in the general case. whenever two or more singular values coincide. and V is n × n. if p − r singular values are zero. If desired. . and write A = Up Σp VpT where Up is m × p. together with the corresponding left singular vectors. ur } . . both the 2-norm and the Frobenius norm m n A F = i=1 j=1 |aij |2 .

and equations are often mutually incompatible because they come from inexact measurements (incompatible linear systems were defined in chapter 2). and its first two equations are incompatible: x1 + x2 cannot be equal to both 1 and √ 3. because measurements are unreliable. A linear system of the form Ax = b (3. but may fail to be unique. the system x1 + x2 x1 + x2 x3 = 1 = 3 = 2 has three unknowns. an exact solution to the system (3. However. This leads to more equations than unknowns (the number m of rows in A is greater than the number n of columns).6) arising from a real-life application may or may not admit a solution. x = [0 2 2]. 3. but it tries to satisfy all of them as closely as possible. THE PSEUDOINVERSE and A are neatly characterized in terms of the SVD: A 2 F A 2 2 2 = σ1 + . which has norm 2 (admittedly. Then. A least-squares solution turns out to be x = [1 1 2]T with residual r = Ax − b = [1 − 1 0]. . Even when m ≤ n the equations can be incompatible. . Often more measurements are available than strictly necessary.6) may not exist. obtained for α = 1. but this is the best we can do for this problem. because of errors in the measurements that produce the entries of A. Incompatibility and underdeterminacy can occur together: the system admits no solution.6) is underdetermined. in the least-squares sense. In other circumstances. or may not be unique. where the double bars henceforth refer to the Euclidean norm. any other vector of the form     1 −1 x =  1  + α 1  2 0 is as good as x. 2 31 = sup x=0 Ax x In the next few sections we introduce fundamental results and applications that testify to the importance of the SVD.3 The Pseudoinverse One of the most important applications of the SVD is the solution of linear systems in the least squares sense. the linear system (3. not enough measurements are available. x cannot exactly satisfy any of the m equations in the system. yields exactly the same residual as x (check this). a vector x that satisfies this equation exactly. it makes more sense to find a vector x that minimizes the norm Ax − b of the residual vector r = Ax − b . For instance. but rank 2. as we learned in chapter 2.and right-hand sides of the equations. Thus. In summary.3. in the sense that it has fewer independent equations than unknowns (its rank r is less than n. this is a rather high residual. In these cases. always exists. For instance. that is. as measured by the sum of the squares of the discrepancies between left. . see again chapter 2). An approximate solution. and the least-squares solution is not unique. in the least-squares sense).3. + σp = σ1 .

This minimum norm solution is the subject of the following theorem. 0 0 ··· . The matrix A† = V Σ† U T 1/σ1 . .  .   .8) because U is an orthogonal matrix. .  .   .1 The minimum-norm least squares solution to a linear system Ax = b..   .  . 0 ··· 0     .. 1/σr 0 . that is.2). Proof. U U = I.32 CHAPTER 3.   . The minimum-norm Least Squares solution to Ax = b is the shortest vector x that minimizes Ax − b that is. . 0 .   . But orthogonal matrices do not change the norm of vectors they are applied to (theorem 3. yn cm 0 0 . U (ΣV T x − U T b) (3.. . THE SINGULAR VALUE DECOMPOSITION If there are several least-squares solutions. . .   . let us spell out the last expression.. In order to find the solution to this minimization problem. so that the last expression above equals ΣV T x − U T b or.  . which both proves uniqueness and provides a recipe for the computation of the solution.  y  r+1  −  cr+1  .1. . then one of them turns out to be shorter than all the others. ··· 0  .  . Theorem 3. Σy − c . x is unique.      0 is called the pseudoinverse of A.  . We want to minimize the norm of the following vector:   σ1 0 ··· 0  y1   c1        0 .    σr yr   cr        . . and is given by ˆ = V Σ† U T b x (3.   .        . with y = V T x and c = U T b. all equally good (or bad).. that is. the shortest vector x that achieves the min Ax − b . . This can be written as T U ΣV T x − b .3. One can therefore redefine what it means to “solve” a linear system so that there is always exactly one solution.7) where       † Σ =     is an n × m diagonal matrix. its norm x is smallest.

In conclusion. we will not be able to zero these differences. .8)). cm  33 and do not depend on the unknown y. . . n .  0− . by imposing the constraint x =1 (3. the norm of Ax − b when x is the solution vector. there is freedom in their choice. the desired y has the following components: yi yi = ci for i = 1. the solution is trivial: the minimum-norm solution to Ax = 0 is x=0. Thus.1 works regardless of the value of the right-hand side vector b. the last n − r components of y are multiplied by zeros. . so they have no effect on the solution. . there is nothing we can do about those differences: if some or all the ci for i = r + 1. and each of them contributes a residual |ci | to the solution. what is desired is a nonzero vector x that satisfies the system (3. Thus. once more. m are nonzero. = yn = 0.4 Least-Squares Solution of a Homogeneous Linear Systems Theorem 3. but it is obviously the one with the smallest norm. When b = 0. this solution is not too interesting. . The residual. yr .  . Although correct. . . Since we look for the minimum-norm solution.9) . . that is. In many applications. the meaning of a least-squares solution is therefore usually modified. on the other hand. we would fall back to x = 0 again. In other words. Without any constraints on x. Of course it is not necessarily the only one (any vector in the null space of A is also a solution. When written as a function of the vector c. by definition). In each of the first r differences. For homogeneous linear systems. .3. that is. . . least-squares solution to the original system is the unique vector ˆ = V y = V Σ+ c = V Σ+ U T b x as promised. this is y = Σ+ c . Thus. In summary. We therefore set yr+1 = . when the system is homogeneous. ∆ 3. . Notice that there is no other choice for y. . x = 0 is the minimum-norm solution to any homogeneous linear system. that is. r σi = 0 for i = r + 1. the square of the residual is m m Ax − b 2 = Σy − c 2 = i=r+1 c2 = i (uT b)2 i i=r+1 which is the projection of the right-hand side vector b onto the complement of the range of A.4. which happens to be an exact solution. for the shortest vector x. because x and y are related by an orthogonal transformation. we also want the shortest y. . since this vector is related to Ax − b by an orthogonal transformation (see equation (3. LEAST-SQUARES SOLUTION OF A HOMOGENEOUS LINEAR SYSTEMS The last m − r differences are of the form  cr+1  . .3.9) as well as possible. the minimum-norm. and minimum-norm solution forces the other entries of y. which is therefore unique: minimum residual forces the choice of y1 . is the norm of Σy − c. . .

on the other case. = yn−k = 0.1. When A is rank deficient. then the minimum-norm solution is unique. The unit-norm Least Squares solution to Ax = 0 is the vector x with x = 1 that minimizes Ax that is. Then.12) is equivalent to equation (3. x =1 Note: when σn is greater than zero the most common case is k = 1. The following theorem provides a recipe for finding this solution. all vectors of the form x = α1 vn−k+1 + .11). . . that is.11) are unit-norm least squares solutions to the homogeneous linear system Ax = 0.10) (3.34 CHAPTER 3. since it is very unlikely that different singular values have exactly the same numerical value. that is. . they achieve the min Ax . that is. . . (3. 2 2 2 2 σ1 y1 + . Proof. .12) From y = V T x we obtain x = V y = y1 v1 + . Since V is orthogonal. ∆ Section 3. it may often have more than one singular value equal to zero. This is obviously achieved by concentrating all the (unit) mass of y where the σs are smallest. vn be the k columns of V whose corresponding singular values are equal to the last singular value σn . this norm is the same as ΣV T x or. . the resulting constrained minimization problem does not necessarily admit a unique solution. the theorem above shows how to express all solutions as a linear combination of the last k columns of V . x = 1 translates to y = 1. . Theorem 3. Furthermore. . . . let vn−k+1 . let k be the largest integer such that σn−k+1 = . x = vn . .10) with α1 = yn−k+1 .5 shows a sample use of theorem 3. Since orthogonal matrices do not change the norm of vectors they are applied to (theorem 3. + yn vn . with y = V T x. Unfortunately. + αk vn with 2 2 α1 + . + σn yn . Σy . . If k > 1. and the unit-norm constraint on y yields equation (3. In any event. . . . . We thus look for the unit-norm vector y that minimizes the norm (squared) of Σy. and shows that there is in general a whole hypersphere of solutions. THE SINGULAR VALUE DECOMPOSITION on the solution. + αk = 1 (3. = σn .1. . that is by letting y1 = . αk = yn . if k = 1.2). so that equation (3. . . The reasoning is very similar to that for the previous theorem.4.1 Let A = U ΣV T be the singular value decomposition of A.4. U ΣV T x . .

15) we must have =0. dm ) and P = (p1 . The best-fit line minimizes the sum of the squared distances. (3. Thus.3). SVD LINE FITTING 35 3.13) where the unit vector n = (a. (3.3. b)T . and the distance between the line n and a point pi is equal to di = |axi + byi − c| = |pT n − c| .5.15) 3. and let be the equation of a line. pm )T . . m .13). (3. . .1 Fitting a Line to a Set of Points ax + by − c = 0 Let pi = (xi . we can assume without loss of generality that n = a2 + b2 = 1 .5.3: The distance between point pi = (xi .5 SVD Line Fitting The Singular Value Decomposition of a matrix yields a simple method for fitting a line to a set of points on the plane. yi )T and line ax + by − c = 0 is |axi + byi − c|. at the minimum (3.14) i pi |c| b a Figure 3.16) If we define the centroid p of all the points pi as p= 1 T P 1.2 The Best Line Fit ∂ d ∂c 2 Since the third line parameter c does not appear in the constraint (3. Thus. . if we let d = (d1 . .15). the line does not change. the best-fit line achieves the min d n =1 In equation (3. If the lefthand side of this equation is multiplied by a nonzero constant. 1 is a vector of m ones. The distance from the line to the origin is |c| (see figure 3. . yi )T be a set of m ≥ 2 points on the plane. orthogonal to the line. is called the line normal. 3. (3. 2 = min P n − c1 n =1 2 . .5.

THE SINGULAR VALUE DECOMPOSITION = ∂ nT P T − c1T (P n − 1c) ∂c ∂ = nT P T P n + c2 1T 1 − 2nT P T c1 ∂c = 2 mc − nT P T 1 = 0 1 T T n P 1. Equivalently. m from which we obtain c= that is. proceed as follows: 1. the residue is min d = σ2 n =1 and more specifically the distances di are given by d = σ2 u2 where u2 is the left singular vector corresponding to σ2 . pm )T . since Qv2 has norm σ2 .16) yields ∂ d ∂c 2 CHAPTER 3.1. we can recall that if n is on a circle. compute the centroid of the points (1 is a vector of m ones): p= 2. to fit a line (a. By replacing this expression into equation (3. In fact. where Q = P − 1pT collects the centered coordinates of the m points. when n = v2 . To summarize. . c = pT n . b. we obtain min d n =1 2 = min P n − 1pT n n =1 2 = min Qn n =1 2 . . We can solve this constrained minimization problem by theorem 3.36 equation (3.4. form the matrix of centered coordinates: Q = P − 1pT 3. the shortest vector of the form Qn is obtained when n is the right singular vector v2 corresponding to the smaller σ2 of the two singular values of Q. Furthermore.15). . compute the SVD of Q: Q = U ΣV T 1 T P 1 m . the SVD 2 Q = U ΣV T = i=1 σi ui vT i yields 2 Qn = Qv2 = i=1 σi ui vT v2 = σ2 u2 i because v1 and v2 are orthonormal vectors. and in order to emphasize the geometric meaning of signular values and vectors. c) to a set of m points pi collected in the m × 2 matrix P = (p1 .

The fitting problem (including fitting a line to a set of points) can be cast either as a maximization or a minimization problem. 5. % the line normal is the second column of V n = V(:. 1). SVD LINE FITTING 4.5. % centroid of all the points p = (P’ * one) / m. An affine subspace is a linear subspace plus a point. error(’matrix P must be m x 2’).3. [U Sigma V] = svd(Q). the residue of the fit is The following matlab code implements the line fitting method. the line normal is the second column of the 2 × 2 matrix V : n = (a. just like an arbitrary line is a line through the origin plus a point. % matrix of centered coordinates Q = P . % assemble the three line coefficients into a column vector l = [n . error(’Need at least two points’). or something close to it. Hint: minimizing the distance between a point and a subspace is equivalent to maximizing the norm of the projection of the point onto the subspace. A useful exercise is to think how this procedure. end if m < 2. Let L be a linear space.one * p’. 2). b)T = v2 . end one = ones(m. Here “plus” means the following. p’ * n]. can be adapted to fit a set of data points in Rm with an affine subspace of given dimension n. the third coefficient of the line is 37 c = pT n min d = σ2 n =1 6. 2). if n ˜= 2. function [l. Then an affine space has the form A = p + L = {a | a = p + l and l ∈ L} . % the smallest singular value of Q % measures the residual fitting error residue = Sigma(2. . residue] = linefit(P) % check input matrix sizes [m n] = size(P).

38 CHAPTER 3. THE SINGULAR VALUE DECOMPOSITION .

Chapter 4

Function Optimization
There are three main reasons why most problems in robotics, vision, and arguably every other science or endeavor take on the form of optimization problems. One is that the desired goal may not be achievable, and so we try to get as close as possible to it. The second reason is that there may be more ways to achieve the goal, and so we can choose one by assigning a quality to all the solutions and selecting the best one. The third reason is that we may not know how to solve the system of equations f(x) = 0, so instead we minimize the norm f(x) , which is a scalar function of the unknown vector x. We have encountered the first two situations when talking about linear systems. The case in which a linear system admits exactly one exact solution is simple but rare. More often, the system at hand is either incompatible (some say overconstrained) or, at the opposite end, underdetermined. In fact, some problems are both, in a sense. While these problems admit no exact solution, they often admit a multitude of approximate solutions. In addition, many problems lead to nonlinear equations. Consider, for instance, the problem of Structure From Motion (SFM) in computer vision. Nonlinear equations describe how points in the world project onto the images taken by cameras at given positions in space. Structure from motion goes the other way around, and attempts to solve these equations: image points are given, and one wants to determine where the points in the world and the cameras are. Because image points come from noisy measurements, they are not exact, and the resulting system is usually incompatible. SFM is then cast as an optimization problem. On the other hand, the exact system (the one with perfect coefficients) is often close to being underdetermined. For instance, the images may be insufficient to recover a certain shape under a certain motion. Then, an additional criterion must be added to define what a “good” solution is. In these cases, the noisy system admits no exact solutions, but has many approximate ones. The term “optimization” is meant to subsume both minimization and maximization. However, maximizing the scalar function f (x) is the same as minimizing −f (x), so we consider optimization and minimization to be essentially synonyms. Usually, one is after global minima. However, global minima are hard to find, since they involve a universal quantifier: x∗ is a global minimum of f if for every other x we have f (x) ≥ f (x∗ ). Global minization techniques like simulated annealing have been proposed, but their convergence properties depend very strongly on the problem at hand. In this chapter, we consider local minimization: we pick a starting point x0 , and we descend in the landscape of f (x) until we cannot go down any further. The bottom of the valley is a local minimum. Local minimization is appropriate if we know how to pick an x0 that is close to x∗ . This occurs frequently in feedback systems. In these systems, we start at a local (or even a global) minimum. The system then evolves and escapes from the minimum. As soon as this occurs, a control signal is generated to bring the system back to the minimum. Because of this immediate reaction, the old minimum can often be used as a starting point x0 when looking for the new minimum, that is, when computing the required control signal. More formally, we reach the correct minimum x∗ as long as the initial point x0 is in the basin of attraction of x∗ , defined as the largest neighborhood of x∗ in which f (x) is convex. Good references for the discussion in this chapter are Matrix Computations, Practical Optimization, and Numerical Recipes in C, all of which are listed with full citations in section 1.4. 39

40

CHAPTER 4. FUNCTION OPTIMIZATION

4.1 Local Minimization and Steepest Descent
Suppose that we want to find a local minimum for the scalar function f of the vector variable x, starting from an initial point x0 . Picking an appropriate x0 is crucial, but also very problem-dependent. We start from x0 , and we go downhill. At every step of the way, we must make the following decisions: • Whether to stop. • In what direction to proceed. • How long a step to take. In fact, most minimization algorithms have the following structure: k=0 while xk is not a minimum compute step direction pk with pk = 1 compute step size αk xk+1 = xk + αk pk k =k+1 end. Different algorithms differ in how each of these instructions is performed. It is intuitively clear that the choice of the step size αk is important. Too small a step leads to slow convergence, or even to lack of convergence altogether. Too large a step causes overshooting, that is, leaping past the solution. The most disastrous consequence of this is that we may leave the basin of attraction, or that we oscillate back and forth with increasing amplitudes, leading to instability. Even when oscillations decrease, they can slow down convergence considerably. What is less obvious is that the best direction of descent is not necessarily, and in fact is quite rarely, the direction of steepest descent, as we now show. Consider a simple but important case, 1 f (x) = c + aT x + xT Qx 2 (4.1)

where Q is a symmetric, positive definite matrix. Positive definite means that for every nonzero x the quantity xT Qx is positive. In this case, the graph of f (x) − c is a plane aT x plus a paraboloid. Of course, if f were this simple, no descent methods would be necessary. In fact the minimum of f can be found by setting its gradient to zero: ∂f = a + Qx = 0 ∂x so that the minimum x∗ is the solution to the linear system Qx = −a . (4.2)

Since Q is positive definite, it is also invertible (why?), and the solution x∗ is unique. However, understanding the behavior of minimization algorithms in this simple case is crucial in order to establish the convergence properties of these algorithms for more general functions. In fact, all smooth functions can be approximated by paraboloids in a sufficiently small neighborhood of any point. Let us therefore assume that we minimize f as given in equation (4.1), and that at every step we choose the direction of steepest descent. In order to simplify the mathematics, we observe that if we let e(x) = ˜ then we have 1 (x − x∗ )T Q(x − x∗ ) 2 (4.3)

1 e(x) = f (x) − c + x∗ T Qx∗ = f (x) − f (x∗ ) ˜ 2

4.1. LOCAL MINIMIZATION AND STEEPEST DESCENT
so that e and f differ only by a constant. In fact, ˜ e(x) = ˜ 1 T 1 1 1 (x Qx + x∗ T Qx∗ − 2xT Qx∗ ) = xT Qx + aT x + x∗ T Qx∗ = f (x) − c + x∗ T Qx∗ 2 2 2 2 1 1 1 f (x∗ ) = c + aT x∗ + x∗ T Qx∗ = c − x∗ T Qx∗ + x∗ T Qx∗ = c − x∗ T Qx∗ . 2 2 2 The result (4.3), e(x) = f (x) − f (x∗ ) , ˜

41

and from equation (4.2) we obtain

is rather interesting in itself. It says that adding a linear term aT x (and a constant c) to a paraboloid 1 xT Qx merely 2 1 shifts the bottom of the paraboloid, both in position (x∗ rather than 0) and value (c − 2 x∗ T Qx∗ rather than zero). Adding the linear term does not “warp” or “tilt” the shape of the paraboloid in any way. Since e is simpler, we consider that we are minimizing e rather than f . In addition, we can let ˜ ˜ y = x − x∗ , that is, we can shift the origin of the domain to x∗ , and study the function e(y) = 1 T y Qy 2

instead of f or e, without loss of generality. We will transform everything back to f and x once we are done. Of ˜ course, by construction, the new minimum is at y∗ = 0 where e reaches a value of zero: e(y∗ ) = e(0) = 0 .

However, we let our steepest descent algorithm find this minimum by starting from the initial point y0 = x0 − x∗ . At every iteration k, the algorithm chooses the direction of steepest descent, which is in the direction pk = − opposite to the gradient of e evaluated at yk : gk = g(yk ) = ∂e = Qyk . ∂y y=y k gk gk

We select for the algorithm the most favorable step size, that is, the one that takes us from yk to the lowest point in the direction of pk . This can be found by differentiating the function e(yk + αpk ) = 1 (y + αpk )T Q(yk + αpk ) 2 k

with respect to α, and setting the derivative to zero to obtain the optimal step αk . We have ∂e(yk + αpk ) = (yk + αpk )T Qpk ∂α and setting this to zero yields αk = − (Qyk )T pk gT p pT p gT g = − Tk k = gk Tk k = gk Tk k . pT Qpk pk Qpk pk Qpk gk Qgk k (4.4)

the basic step of our steepest descent can be written as follows: yk+1 = yk + gk that is.6) so if we can bound the expression in parentheses we have a bound on the rate of convergence of steepest descent. respectively. Introduction to Linear and Nonlinear Programming. FUNCTION OPTIMIZATION gT gk k p gT Qgk k k gT gk k g . . gT Qgk k k (4.42 Thus. AddisonWesley. From the definition of e and from equation (4. To this end. we introduce the following result. often not much. G. The arguments and proofs below are adapted from D. yk+1 = yk − CHAPTER 4. For any vector y there holds 4σ1 σn (yT y)2 ≥ yT Q−1 y yT Qy (σ1 + σn )2 where σ1 and σn are.5) How much closer did this step bring us to the solution y∗ = 0? In other words. yT Qyk gT Qgk k k 2 gT Qgk k gk = Qyk and ⇒ yk = Q−1 gk yT Qyk = gT Q−1 gk k k so that e(yk ) − e(yk+1 ) (gT g )2 = T −1k k T . Luenberger. as we shall now prove. Lemma 4. e(yk ) gk Q gk gk Qgk This can be rewritten as follows by rearranging terms: e(yk+1 ) = 1− (gT gk )2 k T Q−1 g gT Qg gk k k k e(yk ) (4. how much smaller is e(yk+1 ). the largest and smallest singular values of Q. symmetric.1. n × n matrix. relative to the value e(yk ) at the previous step? The answer is. 1973.1 (Kantorovich inequality) Let Q be a positive definite.5) we obtain e(yk ) − e(yk+1 ) e(yk ) = yT Qyk − yT Qyk+1 k k+1 yT Qyk k T gT gk gT gk yT Qyk − yk − gTkQg gk Q yk − gTkQg gk k k k k k yT Qyk k = = = Since Q is invertible we have gT gk gT gk 2 gTkQg gT Qyk − gTkQg k k k k k yT Qyk k 2gT gk gT Qyk − (gT gk )2 k k k .

However.7) must be in the shaded area in figure 4. For the same vector of coefficients θi . n are all on the curve 1/σ. the values 1/σj for j = 1. by the line with ordinate λ(σ) = (σ1 + σn − σ)/(σ1 σn ) .1. . Since 1/σ is a convex function of σ.1. This area is delimited from above by the straight line that connects point (σ1 . with Q symmetric and positive definite. Because Q is positive definite.9) ∂f = a + Qxk ∂x x=xk . For any x0 . the values of φ(σ).7) where the coefficients θi σi . yielding the desired result.4. . . 1/σn ). there are many ways to choose the coefficients θi to obtain a particular value of σ. If we let z = UT y we have 1/ (yT y)2 (yT U T U y)2 (zT z)2 = T = T −1 T = T Q−1 y yT Qy −1 U T y yT U ΣU T y y y UΣ z Σ z z Σz θi = add up to one. each of the singular values σj can be obtained by letting θj = 1 and all other θi to zero. ∆ Thanks to this lemma.1. Let 43 Q = U ΣU T be the singular value decomposition of the symmetric (hence V = U ) matrix Q. Thus an appropriate bound is φ(σ) φ(σ) 1/σ ≥ min = min .8). If we let σ= i=1 n 2 zi z 2 n i=1 θi σi n i=1 θi /σi = φ(σ) ψ(σ) (4. (4. LOCAL MINIMIZATION AND STEEPEST DESCENT Proof. since the smallest of them satisfies σn = min yT Qy > 0 y =1 by the definition of positive definiteness.7) is 1/σ. we can state the main result on the convergence of the method of steepest descent. 1/σ1 ) with point (σn . Theorem 4. ψ(σ).2 Let 1 f (x) = c + aT x + xT Qx 2 be a quadratic function of x. . Of course. the values of the denominator ψ(σ) of (4. The denominator ψ(σ) in (4. Thus. ψ(σ) σ1 ≤σ≤σn λ(σ) σ1 ≤σ≤σn (σ1 + σn − σ)/(σ1 σn ) The minimum is achieved at σ = (σ1 + σn )/2. and λ(σ) are on the vertical line corresponding to the value of σ given by (4.8) then the numerator φ(σ) in (4. that is. all its singular values are strictly positive.7) is a convex combination of points on this curve. the method of steepest descent xk+1 = xk − where gk = g(xk ) = gT gk k g gT Qgk k k (4.

the largest and smallest singular value of Q.3) and (4. Proof.11) (4. converges to the unique minimum point x∗ = −Q−1 a of f .44 CHAPTER 4.λ λ(σ) ψ(σ) φ(σ) σ1 σ2 σ σn σ Figure 4. that xk → x∗ .1: Kantorovich inequality. Since the ratio in the last term is smaller than one. at every step k there holds f (xk+1 ) − f (x∗ ) ≤ σ1 − σn σ1 + σn 2 (f (xk ) − f (x∗ )) where σ1 and σn are. since the minimum of f is unique. By equations (4.12) (f (xk ) − f (x∗ )) . From the definitions y = x − x∗ and e(y) = 1 T y Qy 2 (4. ∆ The ratio κ(Q) = σ1 /σn is called the condition number of Q. respectively. It is easily seen why this happens in the case . The larger the condition number. the closer the fraction (σ1 − σn )/(σ1 + σn ) is to unity.10) we immediately obtain the expression for steepest descent in terms of f and x.6) and the Kantorovich inequality we obtain f (xk+1 ) − f (x∗ ) = e(yk+1 ) = = σ1 − σn σ1 + σn 1− 2 (gT gk )2 k T Q−1 g gT Qg gk k k k e(yk ) ≤ 1− 4σ1 σn (σ1 + σn )2 e(yk ) (4. and the slower convergence. FUNCTION OPTIMIZATION φ.ψ. it follows immediately that f (xk ) − f (x∗ ) → 0 and hence. Furthermore.

and minimization would get there in just one more step. LOCAL MINIMIZATION AND STEEPEST DESCENT 45 x* x0 p0 p1 Figure 4. If β is this limit. which shows the trajectory xk superimposed on a set of isocontours of f (x).2 implies that steepest descent has at best a linear order of convergence. then close to the solution (that is. This is defined as the largest value of q for which the lim xk+1 − x∗ xk − x∗ q k→∞ is finite. in which x is a two-dimensional vector. Thus. In this case. To characterize the speed of convergence of different minimization algorithms. then the first turn would make the new direction point to x∗ . the better.2: Trajectory of steepest descent. that is. the distance of xk from x∗ is reduced by the q-th power at every step. Theorem 4. the residuals |f (xk ) − f (x∗ )| in the values of the function being minimized converge . but very precarious case. for large values of k) we have xk+1 − x∗ ≈ β xk − x∗ q for a minimization method of order q.1. as in figure 4. because then σ1 − σn =0. The minimum of f along that line is tangent to some other. namely. Q is the matrix of second derivatives of f with respect to x. The next step is orthogonal to the latter isocontour (that is. will not pass through x∗ . the analysis above applies once xk gets close enough to the minimum. the farther away a line orthogonal to an isocontour passes from x∗ . For general (that is. the line in the direction pk of steepest descent. the greater the condition number κ(Q). parallel to the gradient). when the starting point x0 is at one apex (tip of either axis) of an isocontour ellipse. σ1 + σn The more elongated the isocontours. non-quadratic) f . In other words. so the higher the order of convergence. lower isocontour. in which κ(Q) = 1. In all other cases. is consistent with our analysis. In summary.2. so that f is well approximated by a paraboloid. which is orthogonal to the isocontour at xk . one iteration will lead to the minimum x∗ .1. In that case. If isocontours were circles (σ1 = σn ) centered at x∗ . and is called the Hessian of f . steepest descent is good for functions that have a well conditioned Hessian near the minimum.4. but can become arbitrarily slow for poorly conditioned Hessians. This case. In fact. and the more steps are required for convergence. at every step the steepest descent trajectory is forced to make a ninety-degree turn. we introduce the notion of the order of convergence. There is one good.

Using line search to find αk guarantees that a minimum in the direction pk is actually reached even when the parabolic approximation is inadequate. u. but there is no general-purpose fix to this problem. In fact. To complete the steepest descent algorithm we need to specify how to check whether a minimum has been reached. One criterion is to check whether the value of f (xk ) has significantly decreased from f (xk−1 ). say. c) otherwise (a. We used Q because it was available. u. if we cannot assume convexity. b. the derivatives of f are close to zero. b. In summary. c) = (a. Let h(α) = f (xk + αpk ) (4. and therefore preferable in most cases. b. The only difficulty here is to find c.4)). we used the Hessian Q in order to compute the optimal step size α (see equation (4. In fact.13) be the scalar function of one variable that is obtained by restricting the function f to the line through the current point xk and in the direction of pk . we can set a = 0. Then. rather than in that of f (x∗ ). c) end end. . b. and then picks a point between a and c. In fact. but its computation during steepest descent would in general be overkill. In fact. c) = (b. A point c that is on the opposite side of the minimum with respect to a can be found by increasing α through values α1 = a. so the function is initially decreasing. In contrast. In our analysis of steepest descent. Since the gradient of f approaches zero when xk tends to x∗ . Shrinking works as follows: if b − a > c − b u = (a + b)/2 if f (u) > f (b) (a. b) end otherwise u = (b + c)/2 if f (u) > f (b) (a. we can set c = αi . so |f (xk ) − f (xk−1 )| may be very small but xk − xk−1 may still be relatively large.46 CHAPTER 4. b = (a + c)/2. b. the steepest descent algorithm can be stopped when xk − xk−1 < where the positive constant is provided by the user. in the sense that a ≤ αk ≤ c. the check on xk is more stringent.13) to the starting point xk . Close to the minimum. FUNCTION OPTIMIZATION linearly. usually one is interested in the value of x∗ . so the minimum must be somewhere between a and c. c) = (a. c) until c − a is smaller than the desired accuracy in determining αk . the Hessian of f (x) requires computing n second derivatives if x is an n-dimensional 2 vector. c) = (u. Thus. Of course. the arguments xk to f can converge to x∗ even more slowly. . only gradient information is necessary to find pk . until h(αi ) is greater than h(αi−1 ). α2 . u) otherwise (a. Line search first determines two points a. if we can assume that h is convex between α1 and αi . . b. corresponding through equation (4. Line search now proceeds by shrinking the bracketing triple (a. the derivative of h at a is negative. we may find the wrong minimum. c that bracket the desired minimum αk . but it is increasing between αi−1 and αi = c. . Here is how line search works. b. and a line search in the direction of pk can be used to determine the step size αk . Another is to check whether xk is significantly different from xk−1 .

This is the idea of Newton’s method. analogously to equation (4. when f (x) is not exactly a paraboloid. iterations are needed. and obtain.2. . and therefore the minimum is somewhere between a and c. Consequently. A minimization algorithm in which the step direction pk and size αk are defined in this manner is called Newton’s method. so line search will find the minimum in a number of steps that is logarithmic in the desired accuracy. the gradient gk is different from a as used in equation (4. NEWTON’S METHOD 47 It is easy to see that in each case the bracketing triple (a.15) is the Newton step. let y(x) = x − Q(x)−1 g(x) be the place reached by a Newton step starting at x (see equation (4. the analysis in the previous section suggests a straight-forward fix to the slow convergence of steepest descent.14) is valid. (4. . Then y(x∗ ) = x∗ because g(x∗ ) = 0. In fact. Newton’s method is quadratic. . a and Q are the coefficients of the Taylor expansion of f around point x = 0. ∂2f ∂x2 n ∂2f ∂xn ∂x1 are the gradient and Hessian of f evaluated at the current point xk . Let 1 f (xk + ∆x) ≈ f (xk ) + gT ∆x + ∆xT Qk ∆x k 2 be the first terms of the Taylor series expansion of f about the current point xk .14) ··· ··· ∂2f ∂x1 ∂xn     x= xk . at every step the interval (a. . Notice that even when f is a paraboloid. The direction is of course undefined once the algorithm has reached a minimum. The greater speed of Newton’s method over steepest descent is borne out by analysis: while steepest descent has a linear order of convergence. while gk and Qk are the coefficients of the Taylor expansion of f around the current point xk .2). c) preserves the property that f (b) ≤ f (a) and f (b) ≤ f (c). gradient and Hessian are constantly reevaluated in Newton’s method. In fact. To the extent that approximation (4. In other words.15)). where gk = g(xk ) = and Qk = Q(xk ) =   ∂2f = T ∂x∂x x=xk  ∂f ∂x x=xk ∂2f ∂x2 1 (4. the new value x1 will be different from x∗ .2) tells us how to jump in one step from the starting point x0 to the minimum x∗ . but convergence can be expected to be faster. In addition.15) whose solution ∆xk = αk pk yields at the same time the step direction pk = ∆xk / ∆xk and the step size αk = ∆xk . b. when αk = 0. and the step defined by equation (4. . 4. c) shrinks to 3/4 of its previous size. which we now summarize.4. and xk+1 − x∗ = y(xk ) − x∗ = y(xk ) − y(x∗ ) . the linear system Qk ∆x = −gk . that is. In fact. .2 Newton’s Method If a function can be well approximated by a paraboloid in the region in which minimization is performed.1). and suppose that at the minimum x∗ the Hessian Q(x∗ ) is nonsingular. The corresponding pk is termed the Newton direction. equation (4. we can set the derivatives of f (xk + ∆x) with respect to ∆x to zero. Of course.

All it can do is to go downhill. Specifically. This quadratic function f (x) was introduced in equation (4. conjugate gradients for the quadratic case can also be seen as an alternative method for solving a linear system.2). The method of conjugate gradients. This paradox. system (4. Thus. Because of the equivalence with a linear system. It will be introe duced in three steps. it will be developed for the simple case of minimizing a quadratic function with positivedefinite and known Hessian. so that x the first term in the right-hand side above vanishes. These high-dimensional problems arise typically from the discretization of partial differential equations. superlinear convergence. or. and the matrix Q is never stored.3 Conjugate Gradients Newton’s method converges faster (quadratically) than steepest descent (linear convergence rate) because it uses more information about the function f being minimized.15) must be solved. as in equation (4. Partial differential equations relate image intensities over space and time to the motion of the underlying image features. However.1). steepest descent takes many steps to converge. the convergence rate of Newton’s method is of order at least two. This means that the number of iterations is fixed.2) of Newton’s method. The conjugate gradients method described in these notes is the so-called Polak-Ribi` re variation. this single iteration in Newton’s method is more expensive. Overall. At every pixel in the image. Storing and inverting a 500. Since y(x∗ ) = x∗ . 4. However. FUNCTION OPTIMIZATION ∂y 1 ∂2y (xk − x∗ ) + xk − x∗ T ∂x x=x∗ 2 ∂x∂xT x=x ˆ 2 where ˆ is some point on the line between x∗ and xk . the first derivatives of y at x∗ are zero. Thus. because it only uses gradient information. The bottom line is that fast convergence requires work that is equivalent to evaluating the Hessian of f . although the version presented here will only work if the matrix of the system is symmetric and positive definite. but it does so by a sequence of n one-dimensional minimizations. while Newton’s method reaches the minimum in one step. is only apparent. We know that in this case minimizing f (x) is equivalent to solving the linear system (4. the method converges to the solution in n steps. Prima facie.48 From the mean-value theorem. and xk+1 − x∗ ≤ c xk − x∗ 2 where c depends on third-order derivatives of f near x∗ . but it only requires gradient information. In contrast. storing and manipulating a Hessian can be prohibitive. 000 unknown motion field values. It effectively solves the linear system (4. called the motion field. This is very important in cases where x has thousands or even millions of components. if an image has. 000 Hessian is out of the question. say. system (4. is motivated by the desire to accelerate convergence with respect to the steepest descent method. Rather than an iterative method. is represented by a vector whose magnitude and direction describe the velocity of the image feature at that pixel. and a line search in the direction pk to find the step size. for a total of n + n derivatives. Say for instance that we want to compute the motion of points in an image as a consequence of camera motion. For a quadratic function. there are n = 500. the work done by conjugate gradients is equivalent to that done by Newton’s method. this motion. 000 × 500. In cases like these. steepest descent requires the gradient gk for selecting the step direction pk . each requiring one gradient computation and one line search. conjugate gradients saves the day. the method of conjugate gradients discussed in this section seems to violate this principle: it achieves fast. but without paying the storage cost of Newton’s method. in which the dimension n of x is thousands or more. Newton’s method approximates f with paraboloids. First.1). because it requires both the gradient gk and the Hessian Qk to be evaluated. similarly to Newton’s method. In 2 addition. we have xk+1 − x∗ = y(xk ) − y(x∗ ) ≤ CHAPTER 4.2) is never constructed explicitly. discussed in the next section. . the Hessian must be inverted. Conjugate gradients works by taking n steps for each of the steps in Newton’s method. and then jumps at every iteration to the lowest point of the current approximation. where n is the number of components of x. at least. a quarter of a million pixels. For very large problems. Steepest descent locally approximates the function with planes. conjugate gradients is a direct method for the quadratic case. however.

4. As we saw in our discussion of steepest descent. . We will see that the conjugate gradients method requires n gradient evaluations and n line searches in lieu of each n × n matrix inversion in Newton’s method. the gradient of f (x). This would then take us to x∗ right away. It is easy to see that in that case n steps will take us to x∗ . and that the quadratic approximation becomes more and more valid as the algorithm converges to the minimum. 1997. We rely instead on the intuition that conjugate gradients solves system (4. with a line search in each direction. and the cost of finding the minimum depends on the desired accuracy. . This combination of fast convergence. . . and low computational cost per iteration explains the popularity of conjugate gradients methods for the optimization of functions of a large number of variables. In fact. the method is no longer direct. this is the main reason for using conjugate gradients. in which the variable x in f (x) is a three-dimensional vector. called isosurfaces. If the function f starts to behave like a quadratic function early. that is. as it was in the quadratic case. the convergence rate of conjugate gradients is superlinear. Then the quadratic function f (x) is constant over ellipsoids. the assumption that the Hessian Q in expression (4. CONJUGATE GRADIENTS 49 Second.3.2). Optimization — Algorithms and consistent approximations.3.1) is known will be removed. Suppose however that we cannot afford to compute this special direction p1 orthogonal to p0 . As a consequence. and we will omit this proof. . positive definite matrix Q. The next step is in the direction of the gradient. modest storage requirements. After n steps we must be done. somewhere between Newton’s method and steepest descent. Given an arbitrary set of ellipsoidal isosurfaces. the second step then yields the minimum in the space spanned by p0 and p1 . centered at the minimum x∗ . will lead to the minimum for spherical isosurfaces. The first step takes from x0 to x1 . When the ellipsoids are spheres. the conjugate gradients method will be extended to general functions f (x). As discussed above. . Springer. In summary. the minimum x∗ is the solution to the linear system Qx = −a . In this case. all the methods we have seen so far involve explicit manipulation of the matrix Q.16) where Q is a symmetric. then we can write 1 1 T x Qx = yT y 2 2 . convergence is fast.17) We know how to solve such a system. so that the new direction p1 is orthogonal to the previous direction p0 . since isosurfaces are spheres. Formal proofs can be found in Elijah Polak. any set of orthogonal directions. How can we start from a point x0 on one of these ellipsoids and reach x∗ by a finite sequence of one-dimensional searches? In connection with steepest descent. as it requires close to the n steps that are necessary in the quadratic case. but that we can only compute some direction p1 orthogonal to p0 (there is an n − 1-dimensional space of such directions!). However. xn . . Consider the case n = 3. on the other hand. pn−1 span the whole space. a certain property that holds in the quadratic case is now valid only approximately.1 The Quadratic Case 1 f (x) = c + aT x + xT Qx 2 Suppose that we want to minimize the quadratic function (4. (4. each line minimization is independent of the others: The first step yields the minimum in the space spanned by p0 . and each of the steps is simple. if f is nearly quadratic in a large neighborhood of the minimum. and the line between x0 and x1 is tangent to an isosurface at x1 . to slow convergence. Finding tight bounds for the convergence rate of conjugate gradients is hard.4. This occurs because the Hessian of f is no longer a constant. but only the quantity gk = Qxk + a that is. this works much better. since p0 . there is a one-to-one mapping with a spherical system: if Q = U ΣU T is the SVD of the symmetric. positive definite matrix. In spite of this. NY. The arguments offered below appeal to intuition. evaluated at n different points x1 . but iterative. we noticed that for poorly conditioned Hessians orthogonal directions lead to many small steps. and so forth. that is. and x has n components. Third. We now consider an alternative solution method that does not need Q.

. Vol 49. if qT qj = 0 . In summary. Res. . Bureau National Standards.4. 409-436. We will henceforth simply say “conjugate” for brevity.j = Σ1/2 U T pi. pn−1 that are mutually conjugate. k k+1 It is somewhat more cumbersome to show that pi and pk+1 for i = 0. So now we need to find a method for generating n conjugate directions without using either Q or its SVD.18) in the algorithm. It is simple to see that pk and pk+1 are conjugate. Of course. In fact. This can be done by induction.19) holds.j becomes pT U Σ1/2 Σ1/2 U T pj = 0 i or pT Qpj = 0 . because Σ and especially U T are too large. which also incorporates the steps from x0 to xn : g0 = g(x0 ) p0 = −g0 for k = 0 . First. pp. i (4. that is. . .19) This condition is called Q-conjugacy. we rewrite these expressions without Q. due to Hestenes and Stiefel (Methods of conjugate gradients for solving linear systems. Here is the procedure. J. pT Qpk+1 k = pT Q(−gk+1 + γk pk ) k = −pT Qgk+1 + k gT Qpk T k+1 p Qpk pT Qpk k k ∂f ∂x x=xk = −pT Qgk+1 + gT Qpk = 0 . so that orthogonality for qi. in the next subsection. 1952). (4. . .2) to produce conjugate rather than orthogonal vectors. if we can find n directions p0 . FUNCTION OPTIMIZATION y = Σ1/2 U T x . We do this in two steps.18) Consequently. . . . i what does this translate into in terms of the directions pi and pj for the ellipsoidal problem? We have qi. there must be a condition for the original problem (in terms of Q) that is equivalent to orthogonality for the spherical problem. section B. we reach the minimum in at most n steps. k are also conjugate. . If two directions qi and qj are orthogonal in the spherical context. then pi and pj are said to be Q-conjugate or Q-orthogonal to each other. . n − 1 αk = arg minα≥0 f (xk + αpk ) xk+1 = xk + αk pk gk+1 = g(xk+1 ) gT Qp k+1 γ k = pT Q p k k k pk+1 = −gk+1 + γk pk end where gk = g(xk ) = is the gradient of f at xk . we cannot use the transformation (4.j . Then.50 where CHAPTER 4. . Details can be found in Polak’s book mentioned earlier. and if we do line minimization along each direction pk . or Q-orthogonality: if equation (4. we find conjugate directions whose definitions do involve Q. The proof is based on the observation that the vectors pk are found by a generalization of Gram-Schmidt (theorem 2.

16) in n steps. n iterations will not suffice. since pk and pk+1 are associated to different Hessians.3. the denominator of γk becomes pT (gk+1 − gk ) = −pT gk = (gk − γk−1 pk−1 )T gk = gT gk . which was constant for the quadratic case. k k k In conclusion. In fact. However. Strictly speaking. while the gradient gk+1 is orthogonal to the isosurface at xk+1 . However. In fact. This expression for γk can be further simplified by noticing that pT gk+1 = 0 k because the line along pk is tangent to an isosurface at xk+1 . the same algorithm can be used. we notice that gk+1 = gk + αk Qpk . and a few cycles of n iterations each will achieve convergence. g(x) = a + Qx so that gk+1 = g(xk+1 ) = g(xk + αk pk ) = a + Q(xk + αk pk ) = gk + αk Qpk . pT gk = 0 . When the function f (x) is arbitrary. We can therefore write γk = gT αk Qpk gT (gk+1 − gk ) gT Qpk k+1 = k+1 = k+1 . the quadratic approximation becomes more and more valid. we obtain the Polak-Ribi` re formula e γk = gT (gk+1 − gk ) k+1 .3 Extension to General Functions We now know how to minimize the quadratic function (4. as the algorithm approaches the minimum x∗ . CONJUGATE GRADIENTS 51 4. gT gk k 4. the Hessian. However.3. now is a function of xk . k−1 Then. or αk Qpk = gk+1 − gk . pT Qpk pT αk Qpk pT (gk+1 − gk ) k k k and Q has disappeared.2 Removing the Hessian The algorithm shown in the previous subsection is a correct conjugate gradients algorithm. we then lose conjugacy. Similarly.4. which is too large. it is computationally inadequate because the expression for γk contains the Hessian Q. To this end. . We now show that γk can be rewritten in terms of the gradient values gk and gk+1 only.3. without ever constructing the Hessian explicitly.

52 CHAPTER 4. FUNCTION OPTIMIZATION .

and x is the derivative of x with respect to t: ˙ x= dx . and one cannot change coordinates for x ˙ without also changing those for x accordingly. the fact that A is square is not a coincidence. one for the domain and one for the range. if V is an orthogonal matrix and we define y = V Tx . that is. dt ˙ In other words. however. the equation b = U ΣV T x can be written as follows U T b = ΣV T x . ˙ ˙ then the definition of x forces us to transform x by V T as well: dx dV Tx ˙ =VT = V Tx . while we need a single transformation. pre-existing relation between x and x. the SVD does nothing useful for systems of linear differential equations. For this equation. In fact. dt dt In brief.Chapter 5 Eigenvalues and Eigenvectors Given a linear transformation b = Ax . Often. we would like to find an orthogonal matrix S and a diagonal matrix Λ such that A = SΛS T 53 (5. In fact. there is an intimate. the singular value decomposition A = U ΣV of A transforms the domain of the transformation via the matrix V T and its range via the matrix U T so that the transformed system is diagonal. T and where Σ is diagonal. of the form ˙ x = Ax where A is n × n. domain and range are intimately related to one another even independently of the transformation A. because it diagonalizes A by two different transformations. x is assumed to be a function ˙ of some real scalar variable t (often time). Ideally. The most important example is perhaps that of a system of linear differential equations. c = Σy where y = V Tx and c = UT b . This is a fundamental transformation to use whenever the domain and the range of A are separate spaces. In fact.1) .

Finally. • the elements on the diagonal of Λ are allowed to be negative. it will be possible to diagonalize A by orthogonal transformations S and S T .1) is not always possible. writing A in the form (5.54 so that if we define CHAPTER 5. if x1 and x2 are mutually orthogonal. 2 = xH S H Sx = xH x = x . the norm squared is defined as x and if S is unitary we have Sx 2 2 = xH x . so unitary generalizes orthogonal for complex matrices. because now we are imposing stronger constraints on the terms of the decomposition.2) . so the Hermitian is a generalization of the transpose. 1 1 In contrast. as well as the inner products between vectors. scalar differential equations. rather than orthogonal. To distinguish invertible from orthogonal matrices we use the symbol Q for invertible and S for orthogonal. in the sense that xH x2 = 0 . for complex matrices we generalize the notion of transpose by introducing the Hermitian operator: The matrix QH (pronounced “Q Hermitian”) is defined to be the complex conjugate of the transpose of Q. rather than real. conjugate transposition becomes simply transposition. in fact. which can be solved separately. This will only occur for very special matrices. If Q happens to be real. we weaken the constraints in several ways: • the elements of S and Λ are allowed to be complex. 1 then Sx1 and Sx2 are orthogonal as well: xH S H Sx2 = xH x2 = 0 . For complex vectors. because it is a system of n independent. EIGENVALUES AND EIGENVECTORS y = ST x we can write the equivalent but diagonal differential system ˙ y = Λy . in the sense that they do not alter the vectors’ norms. If we refer back to figure 3. Furthermore. We will see all of this in greater detail soon.1. they can be even non-real. The solutions can then be recombined through x = Sy . • S is required to be only invertible. now the circle and the ellipse live in the same space. In some cases. A matrix that is equal to its Hermitian is called a Hermitian matrix. a nonunitary transformation Q can change the norms of vectors. In summary. This stands to reason.1) possible. in order to diagonalize a square matrix A from a system of linear differential equations we generally look for a decomposition of A of the form A = QΛQ−1 (5. and the constraint U = V implies that the vectors vi on the circle that map into the axes σi ui of the ellipse are parallel to the axes themselves. Unitary matrices merely rotate or flip vectors. Unfortunately. A matrix S is said to be unitary if S H S = SS H = I . It is like doing an SVD but with the additional constraint U = V . In order to make a decomposition like (5. This is now much easier to handle.

In contrast. diagonal or not.5 1 0. Figure 5. . That real eigenvectors and eigenvalues do not always exist can be clarified by considering the eigenvalue problem from a geometrical point of view in the n = 2 case. The columns of Q are called eigenvectors. and the transformation of B into A (and vice versa) is called a similarity transformation. . are related by A = QBQ−1 .5 0 −0.5) A= 0 2 .3) Thus. 2 1. the columns of qi of Q and the diagonal entries λi of Λ are solutions of the eigenvalue/eigenvector equation Ax = λx . (5. we have derived this equation from the requirement of diagonalizing a matrix by a similarity transformation. Q is invertible. Whenever two matrices A and B. As we know. For some special matrices. λn ) .1 shows the effect of the transformation represented by the matrix √ 2/3 4/ 3 (5. . The equation A = QΛQ−1 can be rewritten as follows: AQ = QΛ or separately for every column of Q as follows: Aqi = λi qi where Q= q1 ··· qn and Λ = diag(λ1 . and Λ is diagonal. The dashed lines are vectors that do not change direction under the transformation. (5. .5 −1 −1. Each point on the unit circle is transformed into some point on the ellipse.5) on a sample of points on the unit circle. they are said to be similar to each other.5 −2 −2 −1 0 1 2 Figure 5.1: Effect of the transformation (5.55 where Q and Λ are complex. and the diagonal entries of Λ are called eigenvalues. this may specialize to A = SΛS H with unitary S. an invertible linear transformation transforms the unit circle into an ellipse.4) which is how eigenvalues and eigenvectors are usually introduced.

Equation (5. In this expression. Therefore. q2 that are mapped into themselves by the original transformation A (see equation (5. Each rotation yields another matrix A. because of the following results. . 5. for system (5. but the resulting ellipse is unchanged. and that a rather diverse array of situations may arise.6) to admit nontrivial solutions. orthonormal eigenvectors. in the sense that det(A − λI) = (−1)n (λ − λ1 ) · . the two eigenvectors are shown as dashed lines. Notice that they do not correspond to the axes of the ellipse. In some cases. square system of equations. by very simple induction. EIGENVALUES AND EIGENVECTORS for a sample of points on the unit circle. perhaps all real. The case of exactly n distinct eigenvalues is of particular interest. the curve-to-curve transformation from circle to ellipse is unique. there are n real.1 can be rotated. as evident from figure 5. defined as the (n − 1) × (n − 1) matrix obtained by removing row i and column j from B. that the left-hand side of equation (5. . pulling the solid lines along. . A square matrix B is rank-deficient iff its determinant. • det(BC) = det(B) det(C).1 Computing Eigenvalues and Eigenvectors Algebraically Let us rewrite the eigenvalue equation Ax = λx as follows: (A − λI)x = 0 . which is called the characteristic equation of A.6) This is a homogeneous. In other words. In figure 5. Given that the directions of the input vectors are generally changed by the transformation A. and the problem is best treated as an algebraic one.3)). · (λ − λn ) where some of the λi may coincide. it is sufficient to recall the following properties from linear algebra: • det(B) = det(B T ).1. . the eigenvalues and their eigenvectors exist. but the point-to-point transformation is not. but they are complex. Bij is the algebraic complement of entry bij . In other words. has n complex solutions. . In particularly good cases. . (5. it is not obvious whether the eigenvalue problem admits a solution at all. The geometric intuition is hidden. and that the coefficient of λn is 1. so x can be assumed to be a unit vector without loss of generality. and that they are not orthogonal.7) is a polynomial of degree n in λ. which admits nontrivial solutions iff the matrix A − λI is rankdeficient.7) From the definition of determinant. In other cases. . det(B) = b11 n i+1 bi1 i=1 (−1) det(Bi1 ) if B is 1 × 1 otherwise is zero. In bad cases.4) is homogeneous in x. Matrices represent point-to-point transformations.1.7). The eigenvalue problem amounts to finding axes q1 . equation (5. bn are linearly dependent. and the matrix A cannot be diagonalized. and we can only triangularize it. (5. bi ··· bj ··· bn ) = − det( b1 ··· bj ··· bi ··· bn ). Thus. just as triangularization was sufficient for solving linear algebraic systems. it follows. Volumes have been written about the properties of the determinant. we have to give up the idea of diagonalizing A. we need det(A − λI) = 0 . In fact. We will see that the answer depends on the matrix A. For our purposes. an n × n matrix has at most n distinct eigenvalues. but not enough eigenvectors can be found. This turns out to be good enough for solving linear differential systems.56 CHAPTER 5. Notice that there are many transformations that map the unit circle into the same ellipse. the circle in figure 5. all eigenvalues exist. • det( b1 • det( b1 ··· ··· bn ) = 0 iff b1 .

. (5. . + ck xk = 0 implies that c1 = .1.2 A Hermitian matrix has real eigenvalues. therefore forcing c2 = 0. .8) For Hermitian matrices (and therefore for real symmetric matrices as well). xk are eigenvectors corresponding to eigenvalues λ1 . we have reduced the summation to one containing k − 1 terms.9) Since A = AH . we obtain xH Ax xH Ax which implies that = = λxH x λ∗ xH x (5. . that the vectors x1 . . .1. . which is still an eigenvector of A: c1 x1 + . the differences in parentheses are all nonzero.9) from the left by xH and equation (5. + ck λk xk = 0 and subtracting this equation from equation (5. the situation is even better. By multiplying by A we obtain c1 Ax1 + . . . we have c1 λ1 x1 + . from c1 x1 + . .5. = ck = 0. that is. . the equation c1 x1 + . = ck = 0. . We can repeat this procedure until only one term remains. Suppose that c1 x1 + . . However. . . . . + ck xk = 0 This entire argument can be repeated for the last equation. . xk corresponding to distinct eigenvalues λ1 . . + ck Axk = 0 and because x1 . . Let λ and x be an eigenvalue of A and a corresponding eigenvector: Ax = λx . the last equation can be rewritten as follows: xH A = λ∗ xH . so that c2 x2 + . . . Thus. By taking the Hermitian we obtain xH AH = λ∗ xH . . . . . . .10) λxH x = λ∗ xH x . + ck λk xk = 0 . Proof. COMPUTING EIGENVALUES AND EIGENVECTORS ALGEBRAICALLY 57 Theorem 5. In summary. + ck−1 (λk−1 − λk )xk−1 = 0 . Proof. + ck xk = 0 we also have c1 λk x1 + . Since all λi are distinct.8) we have c1 (λ1 − λk )x1 + . If we multiply equation (5. and this forces c1 = 0. . . xk are linearly independent. . . . and we can replace each xi by xi = (λi − λk )xi .1 Eigenvectors x1 . We need to show that then c1 = . . λk . + ck xk = 0 where the xi are eigenvectors of a matrix A. + ck−1 xk−1 = 0 . . . λk are linearly independent. Theorem 5. . . ∆ (5.10) from the right by x. . . and so forth. .1. . A matrix A is Hermitian iff A = AH .

∆ Since the two eigenvalues are distinct. Proof. ∆ Theorem 5. we point out a simple but fundamental fact about the eigenvalues of a triangular matrix.4 Eigenvectors corresponding to distinct eigenvalues of a Hermitian matrix are mutually orthogonal. Since the eigenvalue equation Ax = λx is homogeneous in x. so that we have λ = λ∗ as promised. Theorem 5. Notice that the converse is not true: a matrix can have coincident eigenvalues and still admit n independent. and even orthonormal. Let λ and µ be two distinct eigenvalues of A. the scalar xH x is nonzero. the n × n identity matrix has n equal eigenvalues but n orthonormal eigenvectors (which can be chosen in infinitely many ways). If we multiply these two equations by yH from the left and x from the right. First.4.6 The determinant of a triangular matrix is the product of the elements on its diagonal.2 show that when some eigenvalues coincide. λyH x = µyH x (λ − µ)yH x = 0 .1. λ − µ is nonzero.1. and let x and y be corresponding eigenvectors: Ax Ay = = λx µy ⇒ yH A = µyH because A = AH and from theorem 5. . Corollary 5. ∆ Corollary 5.1. respectively. EIGENVALUES AND EIGENVECTORS Since x is an eigenvector. The examples in section 5. For instance. any square matrix with n distinct eigenvalues can be diagonalized by a similarity transformation. the eigenvectors of an n × n Hermitian matrix with n distinct eigenvalues are all mutually orthogonal. rather diverse situations can arise concerning the eigenvectors.3 A real and symmetric matrix has real eigenvalues. Consequently. From theorem 5. A real and symmetric matrix is Hermitian. however.1.1. and we must have yH x = 0. eigenvectors. and any square Hermitian matrix with n distinct eigenvalues can be diagonalized by a unitary similarity transformation.5 An n × n Hermitian matrix with n distinct eigenvalues admits n orthonormal eigenvectors. the vector x can be normalized without violating the equation. ∆ In summary. Proof. the eigenvectors can be made to be orthonormal.1.58 CHAPTER 5. we obtain yH Ax yH Ax which implies or = = λyH x µyH x . Proof.2 µ = µ∗ .

. so is B = A − λI.5. for otherwise we can repeat the argument for the transpose. Matrices with n orthonormal eigenvectors are called normal. Proof.1.. . · (ann − λ) which is equal to zero for λ = a11 . . the matrix 2 0 A= (5. GOOD AND BAD MATRICES 59 Proof. . Normal matrices are good news. . because then the n × n system of differential equations ˙ x = Ax has solution n  x(t) = i=1  ci si eλi t = S  eλ1 t . If A is triangular. The eigenvalues of a matrix A are the solutions of the equation det(A − λI) = 0 .2 Good and Bad Matrices Solving differential equations becomes much easier when matrices have a full set of orthonormal eigenvectors.2. Without loss of generality. ∆ Note that diagonal matrices are triangular. By repeating the argument for B11 and so forth until we are left with a single scalar. which because of the properties above has the same eigenvalues. so this result holds for diagonal matrices as well. For instance. . ann . eλn t   c . ∆ Corollary 5. and from the previous theorem we obtain det(A − λI) = (a11 − λ) · . we can assume a triangular matrix B to be upper-triangular. Then. .11) 0 1 has eigenvalues 2 and 1 and eigenvectors s1 = 1 0 s2 = 0 1 . . 5. .7 The eigenvalues of a triangular matrix are the elements on its diagonal. the only possibly nonzero bi1 of the matrix B is b11 . and the summation in the definition of determinant given above reduces to a single term: det(B) = b11 b11 det(B11 ) if B is 1 × 1 otherwise . This follows immediately from the definition of determinant. we obtain det(B) = b11 · . · bnn .

while 3 1 0 3 has double eigenvalue 3 and only eigenvector [1 0]T . and the vector c of constants ci is c = S H x(0) . and the solution becomes  λt  e 1   −1 . and a full set of eigenvectors may fail to exist. The simplest example of a defective matrix is 0 1 0 0 which has double eigenvalue 0 and only eigenvector [1 0]T . A necessary condition for an n × n matrix to be defective. and there is no q3 . λi are the eigenvalues. not all matrices are as good as these. that is. However.   x(t) = S  eλ1 t . A has to have a repeated eigenvalue if it is to be defective. and the eigenvectors it has are not orthogonal. so zero eigenvalues are not the problem. An example of such a matrix is 2 0 −1 1 √ q2 =   H  S x(0) . but they may not be orthonormal. which has eigenvalues 2 and 1 and nonorthogonal eigenvectors q1 = 1 0 2 2 1 1 . It belongs to the scum of all matrices:   0 2 −1 A= 0 2 1  . . q1 and q2 are not orthogonal to each other. This is conceptually only a slight problem.60 CHAPTER 5. repeated eigenvalues are not a sufficient condition for defectiveness. q2 = 2 0 0 corresponding to eigenvalues 0 and 2 in this order.. 0 0 2 Its eigenvalues are 0. and 2. things can be worse yet. because the unitary matrix S is replaced by an invertible matrix Q. we have seen (theorem 5. because the matrix is singular.. x(t) = Q   Q x(0) . How bad can a matrix be? Here is a matrix that is singular. In fact. there may still be a complete set of n eigenvectors. but independent). Furthermore.1. eλn t Computationally this is more expensive. However. eλn t Fortunately these matrices occur frequently in practice. because a computation of a Hermitian is replaced by a matrix inversion. as we now show. EIGENVALUES AND EIGENVECTORS where S = [s1 · · · sn ] are the eigenvectors. However. repeated twice. . has fewer than n eigenvectors.1) that a matrix with distinct eigenvalues (zero or nonzero does not matter) has a full set of eigenvectors (perhaps nonorthogonal. More compactly. First. as the identity matrix proves. is that it have repeated eigenvalues. Its two eigenvectors are     √ 1 1 2  1 q1 =  0  . . to have fewer than n eigenvectors.

then the eigenvalues of the triangular matrix T are equal to those of the original matrix A.   0  ST x =  . . It is intuitively obvious that any nonzero real vector x can be rotated into a vector parallel to e1 . since it triangularizes any square matrix A by a unitary (possibly complex) transformation: T = S H AS . The eigenvectors. The Schur decomposition does even better. COMPUTING EIGENVALUES AND EIGENVECTORS NUMERICALLY 61 5. and since all the other sj are orthogonal to s1 .   .  .  =  . then that is. so λ is also an eigenvalue for T . T = Q−1 AQ .3. T y = λy where y = Q−1 x .  .1 Rotations into the x1 Axis An important preliminary fact concerns vector rotations. and this product is called the Schur decomposition of A. We will show later on that this allows solving systems of linear differential equations regardless of the structure of the system’s matrix of coefficients. if Ax = λx . Numerically stable and efficient algorithms exist for the Schur decomposition. But the trick is the same: let s1 = x . we have 1    T   T  x s1 s1 x    . This transformation is equivalent to factoring A into the product A = ST S H .  x =  . 5. QT Q−1 x = λx . It may be less obvious that a complex vector x can be transformed into a real vector parallel to e1 by a unitary transformation. however. we will not study these algorithms. Formally. take any orthogonal matrix S whose first column is x s1 = . In fact. x Since sT x = xT x/ x = x . are changed according to the last equation. but only show that all square matrices admit a Schur decomposition. sT sT x n n 0 which is parallel to e1 as desired. these matrices can be triangularized by similarity transformations.  .3. In this note. Let e1 be the first column of the identity matrix.5. as we now show. so some matrices cannot be diagonalized by similarity transformations. x .3 Computing Eigenvalues and Eigenvectors Numerically The examples above have shown that not every n × n matrix admits n independent eigenvectors. Fortunately. It is important to notice that if a matrix A is triangularized by similarity transformations.

a system of linear differential equations can be solved regardless of the structure of the matrix of its coefficients. its eigenvalues can be determined. 0 1 0 . We have sH x = xH x/ x = x .  =  . 0 1 0 . . . .12) . Lemma 5. .   . We are now ready to triangularize an arbitrary square matrix A. if the Schur decomposition of a matrix can be computed.  . .    U H AU      U H AU   0 r 0 .   . .  sH sH x n n x 0 . . such that    T =  λ 0 .   . 0 0              C  . By substituting this into (5.12) and rearranging we have     r r  0   0      AU  . It states that any square matrix can be triangularized by unitary transformations. 0     . Moreover. n × n matrix.3.2 The Schur Decomposition The Schur decomposition theorem is the cornerstone of eigenvalue computations.  x =  .  = λU  . . and unitary transformations preserve eigenvalues. . as we will see later.   .1 If A is an n × n matrix and λ and x are an eigenvalue of A and its corresponding eigenvector. 0 Proof. and 1  sH 1   sH x 1     . Let U be a unitary transformation that transforms the (possibly complex) eigenvector x of A into a real vector on the x1 axis:   r  0    x=U . The diagonal elements of a triangular matrix are its eigenvalues. . 5. . EIGENVALUES AND EIGENVECTORS Now s1 may be complex. .3.62 CHAPTER 5. 0     =      =     λ     λ  r 0 . 0 where r is the nonzero norm of x.   SH x =  . . . Ax = λx then there is a transformation where U is a unitary.  just about like before.  .  T = U H AU (5. Consequently.

S is a unitary matrix. By the inductive hypothesis. 0 where λ is any eigenvalue of A. This is the preferred way to compute eigenvalues numerically. Then from the lemma there exists a unitary U such that   λ  0    U H AU =  .  . Because we can pick any eigenvalue as λ. the order of eigenvalues can be chosen arbitrarily. Partition C into a row vector and an (n − 1) × (n − 1) matrix G: C= wH G . S can be chosen so that the eigenvalues λi of A appear in any order along the diagonal of T . By induction. Suppose it holds for all matrices of order n − 1. COMPUTING EIGENVALUES AND EIGENVECTORS NUMERICALLY    T  1 0 . 0   λ 0 . and S H AS is upper-triangular.  . . Proof. only that it exists. . there is a stable and efficient algorithm to compute the Schur decomposition. ∆ Theorem 5. . S H AS is the Schur decomposition of A.3.2 (Schur) If A is any n × n matrix then there exists a unitary n × n matrix S such that S H AS = T where T is triangular. 0 Clearly. there is a unitary matrix V such that V H GV is a Schur decomposition of G. ∆ This theorem does not say how to compute the Schur decomposition. Let   0 ··· 0 1  0    S=U . 0     . and the corresponding right-hand side is of the form required by the lemma.5. V  . Furthermore. The theorem obviously holds for n = 1: 1A1 = A . . C   .  63      =    The last left-hand side is the first column of T . Since the elements on the diagonal of a triangular matrix are the eigenvalues. .  . Fortunately.3.

but is otherwise unnecessary.1. because AT A is positive semidefinite for every A (lemma 5. a general relation between determinant and eigenvalues. The assumption of distinct eigenvalues made the proof simple.2 (Spectral theorem) Every Hermitian matrix can be diagonalized by a unitary matrix.5). In fact. Theorem 5.4. While this link is very important.4. All that is left to prove is that then its eigenvectors are real. it is useful to remember that eigenvalues/vectors and singular values/vectors are conceptually and factually very distinct entities (recall figure 5. Let now A = ST S H be the Schur decomposition of A. xT Qx = 1 . If we recall that a unitary matrix has determinant 1 or -1. which is both real and rank-deficient. In other words.6 that the determinant of a triangular matrix is the product of the elements on its diagonal. the proof is complete. Furthermore. In fact. Theorem 5. and S H = S T . Λ is real and diagonal. Thus. S is real. First. Positive definite or semidefinite matrices arise in the solution of overconstrained linear systems.64 CHAPTER 5. We already know that Hermitian matrices (and therefore real and symmetric ones) have real eigenvalues (theorem 5. Since A is Hermitian. the Schur decomposition of a Hermitian matrix is in fact a diagonalization.1. and T H = (S H AS)H = S H AH S = S H AS = T .1. that the determinants of S and S H are the same.2).1 The determinant of a matrix is equal to the product of its eigenvalues. In fact. Proof. because 0∗ = 0. ∆ ⇒ A = SΛS H ⇒ A = SΛS T . so Λ is real. T = S H AS. we also establish a link between singular values/vectors and eigenvalues/vectors. and therefore admits nontrivial real solutions. It is positive semidefinite if for every nonzero x we have xT Ax ≥ 0. A = AT In either case. and every real symmetric matrix can be diagonalized by an orthogonal matrix: A = AH A real. now that we have the Schur decomposition. S real . In the process. If in addition the matrix is real. a Hermitian matrix.4 Eigenvalues/Vectors and Singular Values/Vectors In this section we prove a few additional important properties of eigenvalues and eigenvectors. Proof. They also occur in geometry through the equation of an ellipsoid.6). with distinct eigenvalues or not. has real eigenvalues and n orthonormal eigenvectors. Let now A be real and symmetric. But the only way that T can be both triangular and Hermitian is for it to be diagonal. ∆ We saw that an n × n Hermitian matrix with n distinct eigenvalues admits n orthonormal eigenvectors (corollary 5. so are its eigenvectors. We recall that a real matrix A such that for every nonzero x we have xT Ax > 0 is said to be positive definite. and this is the first equation of the theorem (the diagonal of a Hermitian matrix must be real). real or not. we know from theorem 5. Notice that a positive definite matrix is also positive semidefinite. we can state the following stronger result. and that the determinant of a product of matrices is equal to the product of the determinants. The proof is very simple. But eigenvectors are the solution of the homogeneous system (5.5).1). so is T . EIGENVALUES AND EIGENVECTORS 5. given the Schur decomposition. Thus. we know that the eigenvalues of a matrix A are equal to those of the triangular matrix in the Schur decomposition of A.4.

A = SΛS T . symmetric matrix is positive semidefinite iff all its eigenvalues are nonnegative. + cn sn with But then xT Ax = xT A(c1 s1 + . .4. positive definite matrices. + cn λn sn ) = c1 λ1 xT s1 + . positive definite matrices are associated to quadratic forms xT Qx that represent energies or second-order moments of mass or force distributions. . Proof. + cn sn ) = xT (c1 As1 + . and write x = c1 s1 + . + cn Asn ) = xT (c1 λ1 s1 + . The eigenvectors of A are also its singular vectors. ∆ ci = xT si . .4. A is positive definite (semidefinite). . symmetric.4 A real. both left and right. . Since xT Ax > 0 (or ≥ 0) for every nonzero x. then xT Ax ≥ 0 for any nonzero x. Furthermore. . the concepts coincide.4. symmetric.4. we also obtain that if A is real. . symmetric. This result can be used to introduce a less direct link. Theorem 5. i But A = SΛS T with nonnegative diagonal entries in Λ is the singular value decomposition A = U ΣV T of A with Σ = Λ and U = V = S. .3 establishes one connection between eigenvalues/vectors and singular values/vectors: for symmetric. then its eigenvalues are positive. and positive semidefinite. Recall that the eigenvalues in the Schur decomposition can be arranged in any desired order along the diagonal. and in particular sT Asi ≥ 0. sn as an orthonormal basis for Rn . positive semidefinite matrix A are equal to its singular values. but for arbitrary matrices.3 implies one of the two directions: If A is real. from Asi = λsi we obtain sT Asi = sT λsi = λsT si = λ si i i i 2 =λ. .4. If A is positive semidefinite. From the previous theorem. . To this end. we can use these eigenvectors s1 . It is positive definite iff all its eigenvalues are positive. .5. . .4.2). . The following result relates eigenvalues/vectors with singular values/vectors for positive semidefinite matrices. we show that if all eigenvalues λ of a real and symmetric matrix A are positive (nonnegative) then A is positive definite (semidefinite). Since real and symmetric matrices have n orthonormal eigenvectors (theorem 5.4. and positive definite. Conversely. Proof. or at least positive semidefinite (for instance. Theorem 5. then its eigenvalues are nonnegative. . . + cn λn xT sn = λ1 c2 + . + λn c2 > 0 (or ≥ 0) 1 n because the λi are positive (nonnegative) and not all ci can be zero. If the proof of that theorem is repeated with the strict inequality. . Theorem 5. EIGENVALUES/VECTORS AND SINGULAR VALUES/VECTORS 65 in which Q is positive definite. where both Λ and S are real. the entries in Λ are nonnegative. Their physical meaning makes them positive definite.3 The eigenvalues of a real. so that λ ≥ 0.5 AT A is positive semidefinite. let x be any nonzero vector. In physics. energies cannot be negative). Lemma 5. ∆ Theorem 5. In fact.

. AAT = U ΣV T V ΣU T = U Σ2 U T is a Schur decomposition with S = U and T = Λ = Σ2 . Similarly. n. we first need a lemma.8 An n × n matrix is normal if an only if it commutes with its Hermitian: AAH = AH A . This is the definition of a normal matrix. the norm of the i-th row of T is equal to the norm of its i-th column. Bei is the i-th column of B.4. Proof. Let A = ST S H be the Schur decomposition of A. we have AAH = AH A if and only if T T H = T H T . . EIGENVALUES AND EIGENVECTORS 2 ≥ 0. the norm of the i-th row of B equals the norm of its i-th column.4. which is the conjugate of the i-th row of B. If m ≥ n and A = U ΣV T is the SVD of A. the first column of T has norm |t11 |. . Then.13) implies that the i-th column of B has the same norm as the i-th row of B.4. The theorem below characterizes the class of all matrices with this property. then for every i = 1. from the lemma. To prove the theorem. Proof. and the eigenvectors of AAT are the left singular vectors of A. the eigenvalues of AAT are the squares of the singular values of A. (5. a triangular matrix T for which T T H = T H T must be diagonal. then T T H = T H T . AAH = ST S H ST H S H = ST T H S H and AH A = ST H S H ST S H = ST H T S H . ∆ Theorem 5. From BB H = B H B we deduce Bx 2 = xH B H Bx = xH BB H x = B H x 2 . The first row has first entry t11 . Proof. and reason similarly to conclude that T must be diagonal. ∆ We have seen that important classes of matrices admit a full set of orthonormal eigenvectors. that is. ∆ Theorem 5. The converse is also obviously true: if T is diagonal. . For any nonzero x we can write xT AT Ax = Ax CHAPTER 5. for m ≤ n. Similarly. that is. However. for m ≤ n. the equality (5. and B H ei is the i-th column of B H . Because S is invertible (even unitary). We now proceed through i = 2. . . we have AT A = V ΣU T U ΣV T = V Σ2 V T which is in the required format to be a (diagonal) Schur decomposition with S = V and T = Λ = Σ2 . Since conjugation does not change the norm of a vector. the eigenvectors of AT A are the right singular vectors of A. .66 Proof. Thus. the i-th column of the n × n identity matrix. so the only way that its norm can be |t11 | is for all other entries in the first row to be zero. Lemma 5. if and only if A can be diagonalized by a unitary similarity transformation.7 If for an n × n matrix B we have BB H = B H B. ∆ . In fact. Then. AAH = AH A if and only if T is diagonal. Let i = 1. .13) If x = ei . the class of all normal matrices. n.6 The eigenvalues of AT A with m ≥ n are the squares of the singular values of A.

for instance.4.4. Notice that Hermitian (and therefore also real symmetric) matrices commute trivially with their Hermitians. unitary (and therefore also real orthogonal) matrices: UUH = UHU = I . Thus. 67 ∆ Checking that AH A = AAH is much easier than computing eigenvectors.8 is a very useful characterization of normal matrices.8. and orthogonal matrices are all normal. real symmetric. so theorem 5.4. We proved this in the proof of theorem 5. Hermitian.4. . but so do.9 A triangular. EIGENVALUES/VECTORS AND SINGULAR VALUES/VECTORS Corollary 5. unitary. normal matrix must be diagonal. Proof.5.

68 CHAPTER 5. EIGENVALUES AND EIGENVECTORS .

n − 1 . and the vector function b(t) is a function of time. . + c1 + c0 y = b(t) . possibly greater than 1. . dt 69 for i = 0.  . n dt dt dt (6. the coefficients aij in the n × n matrix A are constant.   . we show that scalar differential equations of order greater than one can be reduced to systems of first-order differential equations. These systems have the following form: ˙ x = Ax + b(t) x0 (6.2. dn y dn−1 y dy + cn−1 n−1 + .2).4 and 6. xn n−1 d y dtn−1 With this definition.1) (6. . we recall a general result for the solution of first-order differential systems from the elementary theory of differential equations.   dt  .5. we make this result more specific by showing that the solution to a homogeneous system is a linear combination of exponentials multiplied by polynomials in t.3) In fact. 6. . in which x0 is a known vector. the dot denotes differentiation. in sections 6.2) x(0) = where x = x(t) is an n-dimensional vector function of time t. such an equation can be reduced to a first-order system of the form (6. This result is based on the Schur decomposition introduced in chapter 5. defines the initial value of the solution. The equation (6. Then. in section 6.3. First.1 Scalar Differential Equations of Order Higher than One The first-order system (6. . we have di y dti dn y dtn = = xi+1 dxn . .1) by introducing the n-dimensional vector     y x1  dy   . we set up and solve a particular differential system as an illustrative example.Chapter 6 Ordinary Differential Systems In this chapter we use the theory developed in chapter 5 in order to solve systems of first-order linear differential equations with constant coefficients. = x= . In section 6. . .1) subsumes also the case of a scalar differential equation of order n. Finally. which is numerically preferable to the more commonly used Jordan canonical form.

. 0 −c0 1 0 . the exponential is introduced by other means. n − 1. In some treatments. the exponential is defined through its Taylor series. Ax x0 The two solution components xh and xp can be written by means of the matrix exponential. For matrices. . ORDINARY DIFFERENTIAL SYSTEMS dxi (6. If we write the original system (6.4) dt for i = 1.3) and 0 0 . . .2) is given by x(t) = xh (t) + xp (t) where xh (t) is the solution of the homogeneous system ˙ x = x(0) = and xp (t) is a particular solution of ˙ x = x(0) = Ax + b(t) 0.        ··· 1 · · · −cn−1      . j! Usually1 . .2 General Solution of a Linear Differential System We know from the general theory of differential equations that a general solution of system (6..70 and x satisfies the additional n − 1 equations CHAPTER 6. 0 b(t) 6. . . For the scalar exponential eλt we can write a Taylor series expansion eλt = 1 + λt λ2 t2 + + ··· = 1! 2! ∞ j=0 λj t j . . . 0 −c1     b(t) =    0 1 . 0 0 . j! always. .4). . we obtain the first-order system ˙ x = Ax + b(t) xi+1 = where     A=   0 0 . introduced in the following. and the Taylor series expansion above is proven as a property. .   is the so-called companion matrix of (6. 0 −c2 ··· ··· .1) with initial condition (6. . in calculus classes. . . . the exponential eZ of a matrix Z ∈ Rn×n is instead defined by the infinite series expansion Z Z2 e =I+ + + ··· = 1! 2! Z 1 Not ∞ j=0 Zj .3) together with the n − 1 differential equations (6.

5) deAt dt Thus. It turns out that this infinite sum converges (to an n × n matrix which we write as eZ ) for every matrix Z. and the general term Z j /j! is simply the matrix Z raised to the jth power divided by the scalar j!. since by differentiating this expression for xp we obtain ˙ xp = AeAt 0 t e−As b(s) ds + eAt e−At b(t) = Axp + b(t) . we also know that a particular solution to the nonhomogeneous (b(t) = 0) equation (6.10) xh (t) = eAt x(0) t eA(t−s) b(s) ds . for any vector w. we have the following result. At A2 t2 A3 t3 + + + ··· = 1! 2! 3! ∞ j=0 Aj tj .1) is given by t xp (t) = 0 eA(t−s) b(s) ds .6. (6.9) (6.5) gives deAt dt A2 t A3 t2 + + ··· 1! 2! At A2 t2 = A I+ + + ··· 1! 2! = A+ = AeAt . the function xh (t) = eAt w satisfies the homogeneous differential system ˙ xh = Axh .2).6) is a solution to the differential system (6. GENERAL SOLUTION OF A LINEAR DIFFERENTIAL SYSTEM 71 Here I is the n × n identity matrix. This is easily verified.11) . In summary.1). and xh (t) = eAt x(0) (6. so xp satisfies equation (6. From the elementary theory of differential equations. It can be shown that this solution is unique. Substituting Z = At gives eAt = I + Differentiating both sides of (6.7) (6.1) with b(t) = 0 and initial values (6.2) we obtain w = x0 .8) (6. j! (6.2. The solution to ˙ x = Ax + b(t) with initial value x(0) = x0 is x(t) = xh (t) + xp (t) where and xp (t) = 0 (6. By using the initial values (6.

. pp. we briefly review this solution. Once these two extreme cases (nondefective matrix or all-coincident eigenvalues) have been handled. no. However.3. we will only point out that several methods exist for computing a matrix exponential.12) (6. .3. In the general theory of linear differential systems.3 is superior in practice. the matrix exponential (6. this is shown via the Jordan canonical form. Then. However. Nineteen dubious ways to compute the exponential of a matrix (SIAM Review. 4. If the matrix has a full complement of eigenvectors. and small perturbations in the data can change the results dramatically.5) requires too many terms for a good approximation.7). defective or not. As we have done for the SVD and the Schur decomposition.1. the procedure can be carried out analytically if the functions in the right-hand side vector b(t) can be integrated. a similar result can be found through the Schur decomposition introduced in chapter 5. (6. the method of section 6. since it is based on the numerically sound Schur decomposition. In a fundamental paper on the subject.3 Structure of the Solution For the homogeneous case b(t) = 0. When the matrix A is constant.10) can be written as a linear combination. . we seem to have all we need. 20. . in section 6. ORDINARY DIFFERENTIAL SYSTEMS Since we now have a formula for the general solution to a linear differential system.3. the solution procedure we introduce in section 6. This procedure is based on backsubstitution. 6. This result is briefly reviewed in this section for convenience.1 A is Not Defective In chapter 5 we saw how to find the homogeneous part xh (t) of the solution when A has a full set of n linearly independent eigenvectors. we show a general procedure in section 6. the solution obtained in section 6. consider the first order system of linear differential equations ˙ x = x(0) = Ax x0 . 801-36).3 If A is not defective.3. .1. . λn . . The next section shows how to do this. vol. or is does not. then it has n linearly independent eigenvectors q1 . qn with corresponding eigenvalues λ1 . but we will not discuss how this is done2 . and particularly so about the solution xh (t) to the homogeneous part. as we currently assume. In section 6. Cleve Moler and Charles Van Loan discuss a large number of different methods. expm(A) is the matrix exponential of A. in the paper cited above. of this subsection and of the following one are based on notes written by Scott Cohen. constant matrix A. and produces a result analogous to that obtained via Jordan decomposition for the homogeneous part xh (t) of the solution. we can be much more specific about the structure of the solution (6. we have seen that if (but not only if) A has n distinct eigenvalues then it has n linearly independent eigenvectors (theorem 5.2. we show how to compute the homogeneous solution xh (t) in the extreme case of an n × n matrix A with n coincident eigenvalues. Fortunately. The naive solution to use the definition (6. pointing out that no one of them is appropriate for all situations. 6. A full discussion of this matter is beyond the scope of these notes.1). we do not know how to compute the matrix exponential.3. However. To be sure.3. .13) Two cases arise: either A admits n distinct eigenvalues.2 for the case of n coincident eigenvalues can be applied regardless to how many linearly independent eigenvectors exist.72 CHAPTER 6. with constant vector coefficients. Specifically. of scalar exponentials multiplied by polynomials. For a nonhomogeneous system. . 2 In 3 Parts Matlab.3 for solving a homogeneous or nonhomogeneous differential system for any. Let Q = q1 · · · qn .3. Moler and Van Loan point out that the Jordan form cannot be computed reliably.1.2 is the same as would be obtained with the method of section 6. and we have shown how to find xh (t) by solving an eigenvalue problem.9) of system (6. square.3. we have seen that matrices with coincident eigenvalues can still have a full set of linearly independent eigenvectors (see for instance the identity matrix). However. In chapter 5.

12) is xh (t) = SeΛt S H x(0). 73 (6.3. system (6.16) represents n uncoupled. H . differential equations yi = λi yi . . we see that the solution to the homogeneous system (6. How can we compute the matrix exponential in the extreme case in which A has n coincident eigenvalues. Using the relation x = Qy. and the consequent relation y(0) = Q−1 x(0).12) becomes xh (t) = SeΛt S H x(0). . regardless of the number of its linearly independent eigenvectors? In any case. that is.3.6). Since Aqi = λi qi . then the solution to (6.6. then Q is replaced by the Hermitian matrix S = s1 · · · sn . If A is normal. . Thus we can write T as T = Λ + N. .3. we obtain A = QΛQ−1 . and the solution to (6. (6. . The last equation (6. We recall that S is a unitary matrix and T is upper triangular with the eigenvalues of A on its diagonal. .14) by Q−1 on the right. 6. Similarly. and eS(Λt)S H = SeΛt S H . Comparing with (6. STRUCTURE OF THE SOLUTION This square matrix is invertible because its columns are linearly independent. . ˙ The solution is yh (t) = eΛt y(0). The solution (6. . we derived that the solution to (6. if it has n orthonormal eigenvectors q1 . Multiplying both sides of (6. qn . λn ) is a square diagonal matrix with the eigenvalues of A on its diagonal.2 A Has n Coincident Eigenvalues When A = QΛQ−1 .12) is xh (t) = QeΛt Q−1 x(0).12) can be rewritten as follows: ˙ x = Ax ˙ x = QΛQ−1 x −1 ˙ Q x = ΛQ−1 x ˙ y = Λy.16) where y = Q−1 x.15) Then. it should be the case that −1 eQ(Λt)Q = QeΛt Q−1 . This follows easily from the definition of eZ and the fact that (Q(Λt)Q−1 )j = Q(Λt)j Q−1 . . (6.14) where Λ = diag(λ1 . Q−1 is replaced by S H . where eΛt = diag(eλ1 t . eλn t ). if A = SΛS H . where Λ is diagonal and N is strictly upper triangular. A admits a Schur decomposition A = ST S H (theorem 5.2).6) in this case becomes xh (t) = eS(T t)S x(0) = SeT t S H x(0) = SeΛt+N t S H x(0). . where S is Hermitian. we have AQ = QΛ. homogeneous. .12) is xh (t) = QeΛt Q−1 x(0).

so it remains to show how to compute eN t . We already know how to compute eΛt . and N is strictly upper triangular. then eZ1 +Z2 = eZ1 eZ2 = eZ2 eZ1 . and the exponential reduces to a matrix polynomial. N 2 has two nonzero superdiagonals. we have N j = 0 for all j ≥ n (i. and N 4 is the zero matrix:     0 ∗ ∗ ∗ 0 0 ∗ ∗  0 0 ∗ ∗   0 0 0 ∗  2    N =   0 0 0 ∗ →N = 0 0 0 0 → 0 0 0 0 0 0 0 0     0 0 0 ∗ 0 0 0 0  0 0 0 0   0 0 0 0  4    N3 =   0 0 0 0 →N = 0 0 0 0  . Then N has three nonzero superdiagonals. Λt and N t commute: Λt N t = λ It N t = λt N t = N t λt = N t λ It = N t Λt . for example. in our case.. In fact. for a strictly upper triangular n × n matrix. when all the eigenvalues of A coincide.6) if we can compute eT t = eΛt+N t .12) with initial value (6.e. ORDINARY DIFFERENTIAL SYSTEMS Thus we can compute (6. that N is 4 × 4. This turns out to be almost as easy as computing eΛt when the diagonal matrix Λ is a multiple of the identity matrix: Λ = λI that is.17) where A = S(Λ + N )S H is the Schur decomposition of A. Thus. Therefore. Λ = λI is a multiple of the identity matrix containing the coincident eigenvalues of A on its diagonal. . we can write eΛt+N t = eΛt eN t . N is nilpotent of order n). In summary. It can be shown that if two matrices Z1 and Z2 commute.74 CHAPTER 6. in this case. that is if Z1 Z2 = Z2 Z1 . Suppose. N 3 has one nonzero superdiagonal. The fact that N t is strictly upper triangular makes the computation of this matrix exponential much simpler than for a general matrix Z. the general solution to the homogeneous differential system (6. 0 0 0 0 0 0 0 0 In general.13) when the n × n matrix A has n coincident eigenvalues is given by n−1 xh (t) = SeΛt j=0 N j tj H S x0 j! (6. ∞ n−1 N j tj N j tj eN t = = j! j! j=0 j=0 is simply a finite sum.

.k that contain everything to the right of the corresponding Tii . . . . . .  .20) The triangular matrix T can always be written in the following form:  T11 · · · · · · T1k  0 T22 · · · T2k  T = .i+1 ··· Ti. . with arbitrary b(t). ck (t) where ci has ni entries. . let A = ST S H be the Schur decomposition of A. k are of size ni ×ni (possibly 1×1) and contain all-coincident eigenvalues. .3.. The remaining nonzero blocks Tij with i < j can be in turn bundled into matrices Ri = Ti. yk (t)   and for the initial values  y1 (0)   .18) (6.20) can then be solved by backsubstitution as follows: for i = k down to 1 if i < k di (t) = Ri [yi+1 (t). .      (6.  .  y(t) =  .3 The General Case We are now ready to solve the linear differential system ˙ x = x(0) = Ax + b(t) x0 (6. yk (0) The triangular system (6. j j ni −1 Ni t j=0 j! yi (0) + t 0 eλi I(t−s) j j ni −1 Ni (t−s) j=0 j! (ci (s) + di (s)) ds .21) (6.  .19) in the general case of a constant matrix A. y(0) =   . defective or not. 0 ··· 0 Tkk where the diagonal blocks Tii for i = 1. and the same can be done for  y1 (t)  . In fact. .  c(t) =  .6. . .3. . . STRUCTURE OF THE SOLUTION 75 6. and consider the transformed system ˙ y(t) = T y(t) + c(t) where y(t) = S H x(t) and c(t) = S H b(t) . The vector c(t) can be partitioned correspondingly as follows   c1 (t)  . yk (t)]T else di (t) = 0 (an nk -dimensional vector of zeros) end Tii = λi I + Ni (diagonal and strictly upper-triangular part of Tii ) yi (t) = eλi It end. .

we obtain y(t) = eλIt In scalar form. the solutions to system (6. which can be integrated by parts. triangular case. we have y1 (t) = y1 (0) eλt + t12 y2 (0) t eλt t12 y2 (0) λ2 t (e − eλ1 t ) y1 (t) = y1 (0)eλ1 t + λ2 − λ1 if if t11 = t22 t11 = t22 . This is certainly the case when b(t) is a constant vector b.22) for t11 = t22 and for t11 = t22 have different forms. The second equation of the system. this becomes y1 (t) = y2 (t) = (y1 (0) + t12 y2 (0) t) eλt y2 (0) eλt . (6. Thus. we apply the backsubstitution procedure introduced in this section. since we know that in this case two linearly independent eigenvectors exist (theorem 5. the expression for yi (t) is a direct application of equations (6. and it is easy to verify that this solution satisfies the differential system (6. While y2 (t) is the same in both cases.1. ORDINARY DIFFERENTIAL SYSTEMS In this procedure. we could solve the system by finding the eigenvectors of T . and (6.17) with S = I. = t11 0 1 0 t12 t22 t12 t 1 y1 y2 . because then the integrand is a linear combination of exponentials multiplied by polynomials in t − s.10).22) and the initial value equation y(0) = y0 . y1 ˙ y2 ˙ When t11 = t22 = λ. (6.1).22) y(0) . When t11 = λ1 = t22 = λ2 .11). Instead.76 CHAPTER 6. The solution x(t) for the original system (6. (6. As an illustration. In the general case.22).9). we consider a very small example. d1 (t) = t12 y2 (t) = t12 y2 (0) eλ2 t eλ1 (t−s) d1 (s) ds t 0 = = = = y1 (0)eλ1 t + t12 y2 (0) eλ1 t y1 (0)eλ1 t + t12 y2 (0) eλ1 t 0 e−λ1 s eλ2 s ds e(λ2 −λ1 )s ds t t12 y2 (0) λ1 t (λ2 −λ1 )t e (e − 1) λ2 − λ1 t12 y2 (0) λ2 t y1 (0)eλ1 t + (e − eλ1 t ) λ2 − λ1 y1 (0)eλ1 t + Exercise: verify that this solution satisfies both the differential equation (6. the applicability of this routine depends on whether the integral in the expression for yi (t) can be computed analytically. the 2 × 2 homogeneous. y2 (t) = t22 y2 ˙ has solution We then have and y1 (t) = y1 (0)eλ1 t + 0 t y2 (t) = y2 (0) eλ2 t .18) is then x(t) = Sy(t) . .

1.6. By Hooke’s law. Since forces are proportional to accelerations. Two linear differential equations of the second order4 result. The 4 × 4 matrix of the resulting system has an interesting structure. A CONCRETE EXAMPLE 77 1 v1 rest position of mass 1 1 2 v2 2 rest position of mass 2 3 Figure 6. Because differentiation occurs only with respect to one variable. 6. the two masses would assume the positions indicated by the dashed lines.4. which allows finding eigenvalues and eigenvectors analytically with a little trick. into a real solution in a form appropriate to the problem at hand. Suppose that we want to study the evolution of the system over time. and since accelerations are second derivatives of position. In the absence of external forces. as opposed to partial. is not a problem. The point of this section is to show how to transform the complex formal solution of the differential system. The initial system has two second-order equations. Consider the mechanical system in figure 6. eλt − eλ1 t lim = t eλt . the new equations are differential. these are ordinary differential equations. and is transformed into a first-order system with four equations.4 A Concrete Example In this section we set up and solve a more concrete example of a system of differential equations. because the solution would suddenly switch from one form to the other as the difference between t11 and t22 changes from about zero to exactly zero or viceversa. This would seem to present numerical difficulties when t11 ≈ t22 . We will then transform these into four linear differential equations of the first order. time. . computed with any of the methods described above.1: A system of masses and springs. because of Newton’s law. This. In fact. λ1 →λ λ − λ1 and the transition between the two cases is smooth. the three springs exert forces that are proportional to the springs’ elongations: f1 f2 4 Recall = = c1 v1 c2 (v2 − v1 ) that the order of a differential equation is the highest degree of derivative that appears in it. In the following we write the differential equations that describe this system. however.

The accelerations of masses 1 and 2 (springs are assumed to be massless) are proportional to their accelerations.  (6.25)  u3   v1  ˙ u4 v2 ˙ so that u3 = v1 ˙ while the original system (6. we will first transform it to a system of four first-order equations.26) Likewise. we define four new variables     u1 v1  u   v  u= 2 = 2  (6. ORDINARY DIFFERENTIAL SYSTEMS −c3 v2 where the ci are the positive spring constants (in newtons per meter).24) are replaced by the (known) vector u(0) = In the next section we solve equation (6. ˙ We can now gather these four first-order differential equations into a single system as follows: ˙ u = Au where 0  0  A=  0 0 B 1 0 0 0 0 1 0 0    . and u4 = v2 . As shown in the introduction to this chapter.23) becomes u3 ˙ u4 ˙ =B u1 u2 . We also assume that initial conditions v(0) and ˙ v(0) (6. . which specify positions and velocities of the two masses at time t = 0. according to Newton’s second law: m1 v1 ¨ m2 v2 ¨ or. v(0) ˙ v(0) . To solve the second-order system (6.23) . in matrix form.23). the trick is to introduce variables to denote the first-order derivatives of v. the initial conditions (6. ¨ = Bv v where v= v1 v2 and B= +c − c1m1 2 c2 m2 c2 m1 c2 +c3 − m2 = −f1 + f2 = −c1 v1 + c2 (v2 − v1 ) = −(c1 + c2 )v1 + c2 v2 = −f2 + f3 = −c2 (v2 − v1 ) − c3 v2 = c2 v1 − (c2 + c3 )v2 (6.26). For uniformity.24) are given.78 f3 = CHAPTER 6. so that second-order derivatives of v are first-order derivatives of the new variables.

We then obtain µ1. (6. the eigenvalues of A are the square roots of the eigenvalues of B: if we denote the two eigenvalues of B as µ1 and µ2 . √ √ −α + γ .27) .30) .2 are real and negative. where we recall that A= 0 B I 0 and B= +c − c1m1 2 c2 m2 c2 m1 c2 +c3 − m2 (6.5 Solution of the Example Not all matrices have a full set of linearly independent eigenvectors.1. where γ= α2 − β = 1 4 c1 + c2 c2 + c3 − m1 m2 2 + c2 2 . The eigenvalues of A are solutions to the equation Ax = λx . so that the two solutions µ1.27) can be written as the following pair of equations: z By which yields By = µy with µ = λ2 . We also have that α ≥ γ. λ2 = − −α + γ . If we partition the vector x into its upper and lower halves y and z. z we can write Ax = 0 B I 0 y z = z By so that the eigenvalue equation (6. and I is the 2 × 2 identity matrix. SOLUTION OF THE EXAMPLE 79 6. y x= .2 = −α ± γ . we are lucky. = λy = λz .29) λ1 = √ √ λ3 = −α − γ . λ4 = − −α − γ (6.28) In other words. Here. however. and the four eigenvalues of A. the zeros in A are 2 × 2 matrices of zeros. (6.5. then the eigenvalues of A are √ √ √ √ λ1 = µ1 λ2 = − µ1 λ3 = µ2 λ4 = − µ2 . m1 m2 The constant γ is real because the radicand is nonnegative. With the system of springs in figure 6.6. The eigenvalues µ1 and µ2 of B are the solutions of det(B − µI) = where α= 1 2 c1 + c2 c2 + c3 + m1 m2 and β= c1 c2 + c1 c3 + c2 c3 m1 m2 c1 + c2 +µ m1 c2 + c3 c2 2 +µ − = µ2 + 2αµ + β = 0 m2 m1 m2 are positive constants that depend on the elastic properties of the springs and on the masses.

ORDINARY DIFFERENTIAL SYSTEMS come in nonreal. second-order system (6. Finally.33).34) In these expressions. complex-conjugate pairs.31) the four eigenvectors of A are proportional to the four columns of the following matrix:   c2 c2 c2 c2  Q=  a + λ2 1 c λ1 m21 λ1 a + λ2 1 m1 a + λ2 2 c λ2 m21 λ2 a + λ2 2 a= m1 a + λ2 3 c λ3 m21 λ3 a + λ2 3 m1 a + λ2 4 c λ4 m21 λ4 a + λ2 4 m1    (6. the masses’ motions can be described by the superposition of two sinusoids whose frequencies depend on the physical constants involved (masses and spring constants). In fact. we note that u(t) = QeΛt w . The first equation reads − c2 c1 + c2 − α ± γ y1 + y2 = 0 m1 m1 and is obviously satisfied by any vector of the form y=k c1 +c2 m1 c2 m1 −α±γ where k is an arbitrary constant. the values of λi are given in equations (6. However. As we see in the following. from equation (6.17).31) ±λy The eigenvectors of B are the solutions of (B − (−α ± γ)I)y = 0 . y denotes the two eigenvectors of B. Since we just found four distinct eigenvectors. The amplitudes and phases of the sinusoids. Also the eigenvectors of A can be derived from those of B. and Q is in equation (6. This completes the solution of our system of differential equations.23) can be obtained from equation (6. For k = 0. on the other hand. (6. To simplify our manipulation.26) is then given by equation (6.32) Since ±(−α ± γ) are eigenvalues of B. we can write more simply u(t) = QeΛt Q−1 u(0) where λ1  0 Λ=  0 0  0 λ2 0 0 0 0 λ3 0  0 0   . and from equation (6. m1 The general solution to the first-order differential system (6. depend on the initial conditions.32) must be linearly dependent.28) we see that if y is an eigenvector of B corresponding to eigenvalue µ = λ2 . This is to be expected. the determinant of this equation is zero. the solution to the original.33) where c1 + c2 . it may be useful to add some algebraic manipulation in order to show that the solution is indeed oscillatory. (6. then there are two corresponding eigenvectors for A of the form y x= .80 CHAPTER 6. however.25) by noticing that v is equal to the first two components of u. and the two scalar equations in (6. 0  λ4 (6.30). . since our system of springs obviously exhibits an oscillatory behavior.

6. by using the relation k1 − k2 is purely imaginary k3 − k4 is purely imaginary . the ki must come in complex-conjugate pairs: and (6. v(t) = q1 k1 eλ1 t + k2 e−λ1 t + q2 k3 eλ3 t + k4 e−λ3 t .36) In fact. and derive the general solution v(t) for the original. ejx + e−jx = cos x . We have v(t) = Q(1 : 2. we can write c2 m1 c1 +c2 m1 + λ4 = −λ3 q1 q2 q2 c2 m1 c1 +c2 m1 + q1 λ2 1 and q2 = λ2 3 . :) denotes the first two rows of Q. ∗ k1 = k2 ∗ k3 = k4 . 2 and simple trigonometry we obtain v(t) = q1 A1 cos(ω1 t + φ1 ) + q2 A2 cos(ω2 t + φ2 ) where ω1 = √ α−γ = 1 (a + b) − 2 1 (a + b) + 2 c2 1 2 (a − b)2 + 4 m1 m2 1 c2 2 (a − b)2 + 4 m1 m2 ω2 = √ α+γ = . Since λ2 = −λ1 and (see equations (6.5. second-order problem.35). Since the vectors qi are independent (assuming that the mass c2 is nonzero). Finally.35) We now leave the constants in w unspecified. Numerical values for the constants can be found from the initial conditions u(0) by equation (6. :) = where we defined q1 = Thus. we have v(0) = q1 (k1 + k2 ) + q2 (k3 + k4 ) and from the derivative ˙ v(t) = q1 λ1 k1 eλ1 t − k2 e−λ1 t + q2 λ3 k3 eλ3 t − k4 e−λ3 t we obtain ˙ v(0) = q1 λ1 (k1 − k2 ) + q2 λ3 (k3 − k4 ) . SOLUTION OF THE EXAMPLE where we defined 81 w = Q−1 u(0) . Since the λs are imaginary but v(t) is real. where Q(1 : 2.30)).36) follow. :)eΛt w . (6. this means that k1 + k2 is real k3 + k4 is real from which equations (6. we have Q(1 : 2.

Im denote the real and imaginary part and where the two-argument function arctan2 is defined as follows for (x. on the other hand.25). y) = (0. and not on the initial conditions. This function returns the arctangent of y/x (notice the order of the arguments) in the proper quadrant. φ1 = arctan2 (Im(k1 ). The amplitudes Ai and phases φi . . CHAPTER 6. ORDINARY DIFFERENTIAL SYSTEMS b= c2 + c3 . x) = if x = 0 and y > 0  π  2π  −2 if x = 0 and y < 0 and is undefined for (x. depend on the constants ki as follows: A1 = 2|k1 | . and extends the function by continuity along the y axis. 0)  y if x > 0   arctan( x ) y  π + arctan( x ) if x < 0 arctan2 (y. Re(k1 )) A2 = 2|k3 | φ2 = arctan2 (Im(k3 ).35) and (6. ˙ The two constants k1 and k3 can be found from the given initial conditions v(0) and v(0) from equations (6. 0). Re(k3 )) where Re. y) = (0. m2 Notice that these two frequencies depend only on the configuration of the system.82 and a= c1 + c2 m1 .

To this purpose. One often says that a system evolves over time.1: A general system. studying manipulation requires defining models for how a robot arm can move and for how it interacts with the world. The definition is then illustrated by setting up the dynamic system equations for a simple but realistic application. 83 . Kalman filtering has intimate connections with the theory of algebraic linear systems we have developed in chapters 2 and 3. whose input is the current in a given branch and whose output is a voltage across a pair of nodes. whose inputs are the external temperature. • A chemical reactor. In sections 7. u input S system y output Figure 7. • A mass suspended from a spring.Chapter 7 Stochastic State Estimation Perhaps the most important part of studying a problem in robotics or vision.1 Dynamic Systems In its most general meaning. that of modeling the trajectory of an enemy mortar shell. models take on frequently the form of dynamic systems. When motion is involved. The theory of dynamic systems is both rich and fascinating. As discussed in section 7.1). is to determine a good model for the phenomena and events that are involved. The system reacts to this input and produces an output y (see figure 7. The input is the force applied to the mass and the output is the position of the mass. the theory of state estimation. the term system refers to some physical entity on which some action is performed by means of an input u. A dynamic system is a system whose phenomena occur over time. A dynamic system is a mathematical description of a quantity that evolves over time. in the particular form of Kalman filtering. as is very often the case. 7. For instance. Analyzing image motion implies defining models for how points move in space and how this motion projects onto the image.5. The output can be the temperature of the reaction product. we will develop the theory of the Kalman filter. Simple examples of a dynamic system are the following: • An electric circuit.7.3 through 7. Although in this chapter we will barely scratch its surface. and in section 7. and the supply rate of the gas. as well as in most other sciences. an informal definition of a dynamic system is given in the next section. we will consider one of its most popular and useful aspects. the temperature of the gas being supplied.6 we will see that the shell can be shot down before it hits us.

A description of the system in terms of functions. one can show that there exists a function f such that the state x(t) of the system satisfies the differential equation ˙ x(t) = f (x(t). since the evolution is exactly determined once the initial state x at time 0 is known.2 Uncertainty The systems defined in the previous section are called deterministic. continuous or discrete. knowledge of both x(t1 ) and u[t1 . The relation from state to output. the system is said to be discrete. and the output is a function of the state. In that case. the output y of the system happens to coincide with one of the two state variables. all the quantities in the examples vary continuously with time. Specifically. on the other hand. If time varies continuously. of the state at time t1 and the input over the interval t1 ≤ t < t2 . u(t). For the mass attached to a spring. can be given in the case of a regular system. STOCHASTIC STATE ESTIMATION In all these examples. in a dynamic system the input affects the state.1 State Given a dynamic system. For continuous systems. In other cases. an unrealistic state of affairs. however. and is therefore always deducible from the latter. Thus. In general. which leads to postulating a new quantity. the laws of classical mechanics allow computing the new position and velocity of the mass at time t2 given its position and velocity at time t1 and the forces applied over the interval [t1 . time does not come in quanta. must allow computing the state (and hence the output) at time t2 . which summarizes information about the past and the present of the system. u(·). the laws that govern a given physical . the output is the result of the entire history of the system. Determinism implies that both the evolution function f and the output function h are known exactly. so one cannot compute xk+1 as a function of xk . what is input and what is output is a choice that depends on the application. Also. 7. if time varies discretely. In fact. the modeling problem is to somehow correlate inputs (causes) with outputs (effects). k) where f is some function that represents the change. the system is said to be continuous. t1 . that is. An effort of abstraction is therefore required. the way that the input changes the state at time instant number k into the new state at time instant k + 1 can be represented by a simple equation: xk+1 = f (xk . not just one of its values. In practice. t2 ). Rather. t2 ) where u(·) represents the entire function u. This is. k) . it is more natural to consider time as a discrete variable. uk . The examples above suggest that the output at time t cannot be determined in general by the value assumed by the input quantity at the same point in time. t2 ): x(t2 ) = φ(x(t1 ). in this example. Specifying the initial state x0 completes the definition of a continuous dynamic system. but rather compute x(t2 ) as a functional φ of x(t1 ) and the entire input u over the interval [t1 . all quantities are vectors.1.t2 ) . and k. A discrete dynamic system is completely described by these two equations and an initial state x0 . uk . rather than functionals. as for instance for switching networks and computers. t) where the dot denotes differentiation with respect to time. called the state. Also.1. the state could be the position and velocity of the mass.84 CHAPTER 7. 7. For a discrete system. the value x(t) taken by the state at time t must be sufficient to determine the output at the same point in time. for instance. and uk is the input at time k. is essentially the same as for the discrete case: y(t) = h(x(t). the relation between state and output can be expressed by another function: yk = h(xk . t) . Similarly. Furthermore. differentiable. and with continuous first derivative. for which the functional φ is continuous.

This uncertainty can be represented as noise that perturbs the equations we have presented so far. Then. the state is the (initially unknown) position and velocity of the shell. we want to determine the state x of the system. In particular. t) + ξ(t) . k) + ξk Without loss of generality. is that η and ξ are zero-mean Gaussian random variables with known covariance matrices Q and R. but this is a different story: in general the parameters that enter into the definitions of f and h must be estimated by some method. AN EXAMPLE: THE MORTAR SHELL 85 system are known up to some uncertainty. and so forth. for otherwise the mean can be incorporated into the deterministic part. and the covariance matrix of the output noise ξ is m × m. and the output matrix H is m × n.3 Linearity The mathematics becomes particularly simple when both the evolution function f and the output function h are linear. the noise distributions can be assumed to have zero mean.3 through 7. we consider discretization issues because the physical system is itself continuous. the state propagation matrix F is n × n. component wear. the equations themselves are simple abstractions of a complex reality. the example of the mortar shell will be discussed in order to see some of the technical issues involved in setting up the equations of a dynamic system. The coefficients that appear in the equations are known only approximately. and the mean perturbations are no different.5. Estimating the state then leads to enough knowledge about the shell to allow driving an antiaircraft gun to shoot the shell down in mid-flight. in either f or h. The mean may not be known. In sections 7.5 kilometers higher than your own position. We assume that the input u is a vector in Rp . A common assumption. u(t). = f (xk . respectively. The covariance matrix Q of the system noise η is n × n. A discrete system then takes on the following form: xk+1 yk and for a continuous system ˙ x(t) = y(t) = f (x(t). the system equations become xk+1 yk for the discrete case. and the output y is in Rm . k) + ηk = h(xk .2. uk . which is sometimes valid and always simplifies the mathematics. and ˙ x(t) = y(t) = F (t)x(t) + G(t)u(t) + η(t) H(t)x(t) + ξ(t) = = Fk xk + Gk uk + ηk Hk xk + ξk for the continuous one. that is. You spotted an enemy mortar installation about thirty kilometers away. 7. It is useful to specify the sizes of the matrices involved. while the output is a set of observations made by a tracking system. A more realistic model then allows for some inherent.1. but we choose to model it as a discrete system for easier implementation on a computer. 7. This is a very important task. and can change over time as a result of temperature changes. in the case of the mortar shell.2 An Example: the Mortar Shell In this section. we consider the state estimation problem: given observations of the output y over an interval of time. You want to track incoming projectiles with a Kalman filter so you can aim your guns . on a hill that looks about 0. the input matrix G is n × p. unresolvable uncertainty in both f and h. In fact. Then. t) + η(t) h(x(t). the state x is in Rn .7. For instance.

Because of this. z).3) d when the projectile is at (d.6 kilometers/second for the horizontal component. if t + dt is time instant k + 1 and t is time instant k. and you have little time to get ready. From your high-school physics.2) where g is the gravitational acceleration. equal to 9. Since you are taking measurements every dt = 0. and the size of the blob tells you something about the distance.2. ˙ ˙ Consequently.6 d  d   30     ˆ0 =  x  z  =  0. 7. given that you know the actual size of the projectiles used and all the camera parameters. ˙ 2 The reasoning for the horizontal component d is the same. you want to discretize these equations. Similarly. ˙ 2 since z(0) − gt = z(t).1  ˙ 0. so you just guess some values: 0. you have 1 zk+1 = zk + zk dt − g(dt)2 .2) are continuous. The projectile’s elevation is z e = 1000 (7. You do not know the initial velocity of the projectiles.4) You do not have very precise estimates of the noise that corrupts e and s. so you guess measurement covariances Re = Rs = 1000.2 seconds.86 CHAPTER 7.2) yields z(t + dt) − z(t) 1 1 = z(0) + z(0)(t + dt) − g(t + dt)2 − z(0) + z(0)t − gt2 ˙ ˙ 2 2 1 = (z(0) − gt)dt − g(dt)2 ˙ 2 1 = z(t)dt − g(dt)2 . the aiming angle of the camera gives you elevation information about the projectile’s position.1I4 .1 kilometers/second for the vertical component. you decide to ignore air drag.8 × 10−3 kilometers per second squared. 0). 0. (7. you remember that the laws of motion for a ballistic trajectory are the following: d(t) z(t) ˙ = d(0) + d(0)t 1 = z(0) + z(0)t − gt2 ˙ 2 (7. which you put along the diagonal of a 2 × 2 diagonal measurement covariance matrix R. Since you do not trust your physics much. All you have to track the shells is a camera pointed at the mortar that will rotate so as to keep the projectile at the center of the image. where I4 is the 4 × 4 identity matrix.1) (7. your estimate of the initial state of the projectile is     ˙ −0.5 z where d is the horizontal coordinate. the size of the blob in pixels is s= √ 1000 . Thus. Thus.1 The Dynamic System Equation Equations (7. you are at (0.1) and (7. and dots denote derivatives with respect to time. d2 + z 2 (7. you introduce a state update covariance matrix Q = 0. equation (7. For the z component. except that there is no acceleration: ˙ dk+1 = dk + dk dt . where you see a blob that increases in size as the projectile gets closer.6) (7. STOCHASTIC STATE ESTIMATION accurately. z is the vertical.5) .

7) and (7. ˆk ˆk ˆ dk d d d2 k so that after simplifying we can redefine the measurement to be the discrepancy from the estimated value: ek = ek − 1000 We can proceed similarly for equation (7. We want to replace these equations with linear approximations.7) The two measurements sk and ek just defined can be collected into a single measurement vector yk = sk ek .5) and (7. and the 4 × 1 control matrix G depends on dt. and the two approximate measurement equations (7.7. the 4 × 4 matrix F depends on dt.2 The Measurement Equation The two nonlinear equations (7.2. From the elevation equation (7.4) express the available measurements as a function of the true values of the projectile coordinates d and z. (7.3) and (7. To this end.9) . ˆ ˆ ˆ dk dk d2 k (7. we have zk zk − zk ˆ zk ˆ zk ˆ ˆ ek = 1000 ≈ 1000 + − (dk − dk ) .4): sk = and after simplifying: sk = sk − ˆ 2000 dk zk ˆ ≈ −1000 dk + zk ˆ ˆ ˆ2 + z 2 (d2 + zk )3/2 ˆ2 (d2 + zk )3/2 ˆ2 d ˆ k k .8) 1000 ≈ 2 d2 + zk k 1000 ˆ d2 + zk ˆ2 k − ˆ 1000dk 1000ˆk z ˆ (d − dk ) − (z − zk ) ˆ ˆ2 + z 2 )3/2 k ˆ2 + z 2 )3/2 k (dk ˆk (dk ˆk zk zk ˆ zk ˆ ≈ 1000( − dk ) .3). we develop both equations as Taylor series around the current estimate and truncate them after the linear term. the control scalar u is equal to −g. AN EXAMPLE: THE MORTAR SHELL Equations (7.8) can be written in the matrix form yk = Hk xk where the measurement matrix Hk depends on the current state estimate ˆk : x   ˆ zk ˆ dk 0 0 3/2 3/2 ˆ z2 ˆ z2 (d2 +ˆk ) (d2 +ˆk )  k k Hk = −1000  zk ˆ 0 0 − d1 ˆ2 ˆ d k k (7. −dt2 0 0 dt 1 2 7.2. The two matrices F and G are as follows:     0 1 0 0 0  dt 1 0 0   0    F = G=  0 0 1 0   dt  .6) can be rewritten as a single system update equation xk+1 = F xk + Gu where  ˙ dk  d  xk =  k   zk  ˙ zk  87 is the state.

we introduce the Kalman filter from the simpler point of view of least squares estimation. knowing x0 instead is equivalent. we would like to know where the mortar shell is right now. Given a discrete dynamic system xk+1 yk = = Fk xk + Gk uk + ηk Hk xk + ξk (7.3 State Estimation In this section. and T. In other words. . A classical example of this situation in computer vision is the reconstruction of three-dimensional shape from a sequence of images. at least when the dynamics of the system are known exactly (the system noise ηk is zero). so by itself it conveys no three-dimensional information. It is really the ensemble of all observations that let one estimate the state. since we have developed all the necessary tools in the first part of this course. the estimation problem is defined in some more detail. thereby determining xk as well. 7. . Thus. we frantically start studying state estimation. . the hat x means that the quantity is an estimate. . . So what else is there left to do? From the observations. Even without noise. the second to which measurements are being used for the estimate. Then. E. and section 7. The original paper by Kalman (R.” Transactions of the ASME Journal Basic Engineering. Qk ) N (0. S. one observation happens to be sufficient). “Kalman filter-based algorithms for estimating depth from image sequences. in general. the state of the dynamic system. Kalman. . Matthies. Broida. and in particular Kalman filtering. and perhaps predict where it will be in a few seconds. “A new approach to linear filtering and prediction problems. ˆ as well as a (possibly completely wrong) estimate x0 of the initial state and an initial covariance matrix P0 of the estimate ˆ0 . July 1990. Also. The next section defines the state estimation problem for a discrete dynamic system in more detail. 82:34–45. and yet observations are processed one at a time. Section 7. A scheme that refines an initial estimation of the state as new observations are acquired is called a recursive1 state estimation system. the Kalman filter computes the optimal estimate ˆk|k at time k given the measurements y0 . Kanade. Clearly. Chandrashekhar. A single image is two-dimensional. 1 The term “recursive” in the systems theory literature corresponds loosely to “incremental” or “iterative” in computer science. STOCHASTIC STATE ESTIMATION As the shell approaches us. Rk ) . section 7. The next few sections will be read under this impending threat. September 1989. from x0 we can simulate the system up until time t. T.7 establishes a connection between the Kalman filter and the solution of a linear system.6 reconsiders the example of the mortar shell. . the first k in the subscript refers to which variable is being estimated. and R. In these expressions. . as they become available.J.” International Journal of Computer Vision. The Kalman filter is one of the most versatile schemes for recursive state estimations. Szeliski.5 develops the equation of the Kalman filter. and R.10) (7. ηk ξk ∼ ∼ N (0.11) where the system noise ηk and the measurement noise ξk are Gaussian variables. See for instance L. Finally. yj . In fact. Kalman filters exist that recover shape information from a sequence of images. 26(4):639–656. we do not want to have all the observations before we shoot: we would be dead by then. Chellappa. 3(3):209-236. This is a very interesting aspect of state estimation. we want to know xk . “Recursive 3-D motion estimation from a monocular image sequence. in the hope to build a system that lets us shoot down the shell before it hits us. section 7. ˆi|j is the estimate of the value that x x assumes at time i given the first j + 1 measurements y0 . Most importantly.4 defines the essential notions of estimation theory that are necessary to understand the quantitative aspects of Kalman filtering. Here. Knowing the model for the mortar shell amounts to knowing the laws by which the object moves and those that relate the position of the projectile to our observations. yk .” IEEE Transactions on Aerospace and Electronic Systems. 1960) is still one of the most readable treatments of this subject from the point of view of stochastic estimation.88 CHAPTER 7. a single observation yk may not be sufficient to determine the state xk (in the example. so we can direct an antiaircraft gun to shoot down the target. The x x filter also computes an estimate Pk|k of the covariance of ˆk|k given those measurements. .

Thus. in the following sense. 7. Propagation then accounts for the evolution of the system state. while that about the measurements may remain the same. by how much should we correct our estimate of the state? We do not want to make ˆk|k y coincide with yk . Then. At time k. .1 Update The covariance matrix Pk|k must be computed in order to keep the Kalman filter running. and we can consider the residue rk = yk − ˆk|k−1 = yk − Hk ˆk|k−1 . the estimate x ˆk|k−1 = Hk ˆk|k−1 y x is still our best bet. yk−1 .3. as well as the consequent growing uncertainty. just before the new measurement yk comes in.3.11). . At the same time. . through the measurement equation (7. The update stage of the Kalman filter uses Rk and Pk|k−1 to weigh past evidence (ˆk|k−1 ) and new observations (yk ). Now yk becomes available. Even if ˆk|k−1 is not exact. . Now we face the problem of incorporating the new measurement yk into our estimate. The question however is. STATE ESTIMATION 89 yk-1 k-1 Hk ^ xk-1 | k-1 Pk-1 | k-1 propagate ^ xk | k-1 Pk | k-1 yk k ^ yk | k-1 ^ xk | k Pk | k ^ xk+1 | k Pk+1 | k yk+1 time k+1 update propagate Figure 7. this uncertainty becomes usually smaller: Pk|k < Pk|k−1 . The idea is that as time goes by the uncertainty on the state decreases. .2. That would mean that we trust the new measurement completely. The uncertainty about the state just before the new measurement yk becomes available is Pk|k−1 . so that the new prediction ˆk|k = Hk ˆk|k y x of the measurement value is closer to the measurement yk than the old prediction ˆk|k−1 = Hk ˆk|k−1 y x that we made just before the new measurement yk was available. that is. This stage is represented graphically in the middle x of figure 7. also the uncertainty measure Pk|k−1 must be updated. we x need some criterion for comparing the quality of the new measurement yk with that of our old estimate ˆk|k−1 of the x state.7. Because a new measurement has been read. of transforming ˆk|k−1 into ˆk|k . the covariance of the observation error.2: The update stage of the Kalman filter changes the estimate of the current system state xk to make the prediction of the measurement closer to the actual measurement yk . so that it becomes available for the next step. we have an estimate ˆk|k−1 of the state vector xk based on the x previous measurements y0 . we probably need to correct our estimate of the state xk . measurements count less and less as the estimate approaches its true value. The uncertainty about the former is Rk . even if the latter was obtained through a large number of previous measurements. If ˆk|k−1 were exact. y x If this residue is nonzero. we could compute the new measurement yk x x x without even looking at it. but that we do not trust our state estimate ˆk|k−1 at all.

in this somewhat trivial sense. x In summary. the covariance matrix Pk|k contains all the necessary information about the probabilistic part of the system. x Hopefully. x because of the system noise ηk . that is. so that the system (7. The system equation (7. and inaccuracies in our model of how this happens lead to greater uncertainty. But between time k and time k + 1 both state and covariance may change. 7. In this case. . then L is the inverse of h.2. for a linear observation function. . a problem of estimation. this intuitive introduction to Kalman filtering gives you an idea of what the filter does. In particular. starting with the definition of a linear estimation problem. and then considering the attributes “best” and “unbiased” in turn.3. no matter how the phrase “as close as possible” is interpreted. .4 BLUE Estimators In what sense does the Kalman filter use covariance information to produce better estimates of the state? As we will se later. Thus. The optimal estimator is then represented by the matrix L = H −1 and the optimal estimate is ˆ = Ly . in that case ˆ is equal to x. The Kalman filter itself is derived in section 7. In fact.4. (7. This increase in uncertainty depends on the system noise covariance Qk . Inverting a function is an example of estimation. x A less trivial example occurs.1 Linear Estimation Given a quantity y (the observation) that is a known function of another (deterministic but unknown) quantity x (the state) plus some amount of noise.13) is. we see what this means. If the function h is invertible and the noise term n is zero. we have so far considered the criterion x that prefers a particular ˆ if it makes the Euclidean norm of the vector y − Hx as small as possible.2 Propagation Just after arrival of the measurement yk . In this section. The state changes according to the system equation (7. For linear systems. and again one must say in what sense ˆ is required to be “as close as possible” to x. To turn these concepts into a quantitative algorithm we need some preliminaries on optimal estimation. just as the state vector xk represents all the information necessary to describe the evolution of a deterministic system. about how both the system noise ηk and the measurement noise ξk corrupt the quality of the state estimate ˆk|k . . Similarly. our uncertainty about this estimate may be somewhat greater than one time epoch ago. nonsingular system x x y = Hx (7.90 CHAPTER 7.10) essentially “dead reckons” the new state from the old. and what information it needs to keep working.5. This is the x . Both these changes are shown on the right in figure 7.10).13) is overconstrained. yk should reflect this change as well. both state estimate and covariance must be propagated to the new time k + 1 to yield the new state estimate ˆk+1|k and the new covariance Pk+1|k .12) the estimation problem amounts to finding a function ˆ = L(y) x such that ˆ is as close as possible to x. y = h(x) + n . the Kalman filter computes the Best Linear Unbiased Estimate (BLUE) of the state. 7. when the matrix H has more rows than columns. STOCHASTIC STATE ESTIMATION 7. there is usually no inverse to H. so our estimate ˆk+1|k of xk+1 given y0 . and its value ˆ given the observations y x x is called an estimate. both state estimate and state covariance matrix have been updated as described above. The function L is called an estimator. which are discussed in the next section. and any distance between ˆ and x must be zero. solving a square.

4. Even equation (7.12) with h a linear function and n Gaussian zero-mean noise with the indentity matrix for covariance. In the least squares approach to solving a linear system like (7. (7. In fact. so requiring equality is hopeless. x 2 = nT n (the square is of course irrelevant when it comes to minimization). Thus.2 Best In order to define what is meant by a “best” estimator. even. However. we have W n = W (y − Hx) = W y − W Hx so we are simply solving the system W y = W Hx in the traditional. We know the solution from normal equations: ˆ = ((W H)T W H)−1 (W H)T W y = (H T R−1 H)−1 H T R−1 y .13).14) does not change the nature of the problem.4. evaluated at the solution ˆ. alternatively.4. Notice that the observation function h can still be nonlinear. In section 7. an overconstrained system of the form (7. “unweighted” sense. each equation counts the same when computing the norm of the residue. An estimator is said to be linear if the function L is linear. is only slightly different from its unweighted version. Replacing (7. the Euclidean norm of the residue (7.2.15) and we would like to make that residue as small as possible in the sense of the Euclidean norm. we will see that in a very precise sense ordinary least squares solve a particular type of estimation problem. The noise term. the estimation problem for the observation equation (7. The best estimator for a linear observation function happens to be a linear estimator. we now Wn 2 = nT R−1 n . this can be enforced by some scaling of the entries of n or. by some linear transformation of them: n → Wn so instead of minimizing n minimize where R−1 = W T W is a symmetric. we will probably have an estimator that produces a worse estimate than a nonlinear one. In fact.14) are really the same problem. If L is required to be linear but h is not. What the least squares approach is really saying is that even at the solution ˆ there is some residue x n = y − Hˆ x (7. one needs to define a measure of goodness of an estimate. This amounts to saying that we have reasons to prefer the quality of some equations over others or.13) has no exact solution when there are more independent equations than unknowns.15) treats all components (all equations in (7. In other words. 7. can be used to generalize the problem. In fact. that we want to enforce different equations to different degrees. if the equality sign is to be taken literally.13) by a “noisy x equation”. This minimization problem. However. From the point of view of least squares. BLUE ESTIMATORS 91 (unweighted) least squares criterion. it still makes sense to look for the best possible linear estimator. y = Hx + n (7.7.14) is the correct version. however.13) and its “noisy” version (7. different equations can have noise terms of different variance. called weighted least squares. nonnegative-definite matrix. this distance is defined as the Euclidean norm of the residue vector y − Hˆ x between the left and the right-hand sides of equation (7. namely.13).14)) equally.

92 CHAPTER 7. Then the linear estimator L is unbiased if an only if LH = I . Some matrix norms were mentioned in section 3. ˆ is a function of a random vector (noise). STOCHASTIC STATE ESTIMATION Interestingly. Lemma 7.1 Let n in equation (7. from measurements corrupted by different noise samples from the same distribution.4. the identity matrix. we also want it to be unbiased.4. First.4 The BLUE ˆ = Ly x We now address the problem of finding the Best Linear Unbiased Estimator (BLUE) of x given that y depends on x according to the model (7. which in turn are corrupted by noise. We consider this different approach because it will let us show that the Kalman filter is optimal in a very useful sense. Thus. 7. Formally. x x Minimizing a matrix. Mathematically. this same solution is obtained from a completely different criterion of goodness of a solution ˆ. we only use properties shared by all norms. if we estimate the same quantity many times. Intuitively. In this sense. in the derivations that follow. however. It makes therefore sense to measure the quality of an estimator by requiring that its variance be as small as possible: the fluctuations of the estimate ˆ with respect to the true (unknown) value x from one estimation experiment to the x next should be as small as possible.14). in the sense that if repeat the same estimation experiment many times we neither consistently overestimate nor consistently underestimate x. we want to choose a linear estimator L such that the estimates ˆ = Ly x it produces minimize the following covariance matrix: P = E[(x − ˆ)(x − ˆ)T ] . we give a necessary and sufficient condition for L to be unbiased.16) be zero mean. and is therefore x a random vector itself.3 Unbiased In additionto requiring our estimator to be linear and with minimum covariance. E[x − ˆ] x = E[x − Ly] = E[x − L(Hx + n)] = E[(I − LH)x] − E[Ln] = (I − HL)E[x] (7. This x criterion is a probabilistic one. which is repeated here for convenience: y = Hx + n . most interesting matrix norms are equivalent. this translates into the following requirement: E[x − ˆ] = 0 x and E[ˆ] = E[x] . The new criterion is the so-called minimum-covariance criterion.16) . The estimate ˆ of x is some function of the x measurements y. we can pick any norm we like. In fact. we obtain different estimates.4. so which norm we actually use is irrelevant. requires a notion of “size” for matrices: how large is P ? Fortunately. x 7. Proof. Thus.2. the estimates are random. in the sense that given two different definitions P 1 and P 2 of matrix norm there exist two positive scalars α. β such that α P 1 < P 2 <β P 1 .

we simply substitute L0 for L in P = LRLT : x P as promised. let L be any (other) linear unbiased estimator.18) is the best. In conclusion.7. To prove that the covariance P of ˆ is given by equation (7. 0 0 RLT = RR−1 H(H T R−1 H)−1 = H(H T R−1 H)−1 0 From (7. For this to hold for all x we need I − LH = 0.4. when L = L0 . so this expression is minimized when the second term vanishes. We can trivially write and P = = LRLT = [L0 + (L − L0 )]R[L0 + (L − L0 )]T L0 RLT + (L − L0 )RLT + L0 R(L − L0 )T + (L − L0 )R(L − L0 )T . = L0 RLT = (H T R−1 H)−1 H T R−1 RR−1 H(H T R−1 H)−1 0 = (H T R−1 H)−1 H T R−1 H(H T R−1 H)−1 = (H T R−1 H)−1 ∆ . BLUE ESTIMATORS since E[Ln] = L E[n] and E[n] = 0. 0 But L and L0 are unbiased. so it is zero as well. x x Proof. 0 the sum of two positive definite or at least semidefinite matrices.4.17) because L is unbiased. that it has minimum covariance. 0 (L − L0 )RLT = 0 . This proves that the estimator given by (7.18) is the best choice. Theorem 7. We can write P = = = E[(x − ˆ)(x − ˆ)T ] = E[(x − Ly)(x − Ly)T ] x x E[(x − LHx − Ln)(x − LHx − Ln)T ] = E[((I − LH)x − Ln)((I − LH)x − Ln)T ] E[LnnT LT ] = L E[nnT ] LT = LRLT (7. so that LH = I. To show that L0 = (H T R−1 H)−1 H T R−1 L = L0 + (L − L0 ) (7. P = L0 RLT + (L − L0 )R(L − L0 )T . so LH = L0 H = I. 93 ∆ And now the main result.17). and The term L0 R(L − L0 )T is the transpose of this. For such matrices. that is. that is.18) we obtain so that (L − L0 )RLT = (L − L0 )H(H T R−1 H)−1 = (LH − L0 H)(H T R−1 H)−1 .2 The Best Linear Unbiased Estimator (BLUE) ˆ = Ly x for the measurement model y = Hx + n where the noise vector n has zero mean and covariance R is given by L = (H T R−1 H)−1 H T R−1 and the covariance of the estimate ˆ is x P = E[(x − ˆ)(x − ˆ)T ] = (H T R−1 H)−1 . the norm of the sum is greater or equal to either norm.

which is related to the state xk by the equation yk = Hk xk + ξk with error covariance T E[ξk ξk ] = Rk . and packaging the error covariances into a single. linear. Qk ) N (0. The filter also x computes the covariance Pk|k of the error ˆk|k − xk given those measurements. block-diagonal matrix. We now apply the results from optimal estimation to the problem of updating and propagating the state estimates and their error covariances. two pieces of data are available. Rk ) . . we have y = Hxk + n . . yk . ˆ as well as the best. Another way of saying this is that the estimate ˆk|k−1 differs from the true state xk by an error term ek whose covariance is Pk|k−1 : x ˆk|k−1 = xk + ek x with E[ek eT ] = Pk|k−1 . unbiased estimate ˆk|k at time k given the measurements y0 . 7.19) where the system noise ηk and the measurement noise ξk are Gaussian random vectors.5 The Kalman Filter: Derivation We now have all the components necessary to write the equations for the Kalman filter. One is the estimate ˆk|k−1 of the state xk given measurements up to but not x including yk .2. k (7. n ∼ N (0. Given a dynamic system with system and measurement equations xk+1 yk = = Fk xk + Gk uk + ηk Hk xk + ξk (7.20) The other piece of data is the new measurement yk itself. given a linear measurement equation y = Hx + n where n is a Gaussian random vector with zero mean and covariance matrix R. (7. the Kalman filter computes the best. STOCHASTIC STATE ESTIMATION 7. Computation occurs according to the x phases of update and propagation illustrated in figure 7.1 Update At time k. This estimate comes with its covariance matrix Pk|k−1 . To summarize. ηk ξk ∼ ∼ N (0. linear. unbiased estimate x0 of the initial state with an error covariance matrix P0 . . R) .21 into one. the best linear unbiased estimate ˆ of x is x where the matrix ˆ = P H T R−1 y x P = E[(ˆ − x)(ˆ − x)T ] = (H T R−1 H)−1 x x ∆ is the covariance of the estimation error.20 and 7.5.21) We can summarize this available information by grouping equations 7. .94 CHAPTER 7. Thus.

95 As we know. These expressions are somewhat wasteful. unbiasedness requires ˆk+1|k = Fk ˆk|k + Gk uk . ˆk|k−1 x yk .5. these two update equations are now rewritten in a more efficient and more familiar form. x 7. We have −1 Pk|k = H T R−1 H = I T Hk −1 Pk|k−1 0 0 −1 Rk I Hk −1 T −1 = Pk|k−1 + Hk Rk Hk and ˆk|k x = = = = = In the last line.19. and the matrix x T −1 Kk = Pk|k Hk Rk ∆ ∆ Pk|k H T R−1 y Pk|k −1 Pk|k−1 T −1 Hk Rk ˆk|k−1 x yk −1 T −1 Pk|k (Pk|k−1 ˆk|k−1 + Hk Rk yk ) x −1 T −1 T −1 Pk|k ((Pk|k − Hk Rk Hk )ˆk|k−1 + Hk Rk yk ) x T −1 ˆk|k−1 + Pk|k Hk Rk (yk − Hk ˆk|k−1 ) . because it specifies the amount by which the residue must be multiplied (or amplified) to obtain the correction term that transforms the old estimate ˆk|k−1 of the state xk into its new x estimate ˆk|k . H= I Hk . For this reason. and the noise term ηk is zero mean.7. because the matrices H and R contain many zeros.2 Propagation Propagation is even simpler. THE KALMAN FILTER: DERIVATION where y= and where n has covariance R= Pk|k−1 0 0 Rk .5. the difference rk = yk − Hk ˆk|k−1 x is the residue between the actual measurement yk and its best estimate based on ˆk|k−1 . the solution to this classical estimation problem is ˆk|k x Pk|k = = Pk|k H T R−1 y (H T R−1 H)−1 . x x is usually referred to as the Kalman gain matrix. x x . This pair of equations represents the update stage of the Kalman filter. n= ek ξk . Since the new state is related to the old through the system equation 7.

Notice that the covariance goes to zero only asymptotically. we now show that estimating the initial state x0 from the first k + 1 measurements. x 7. The basic recurrence equations (7.10) and (7. For this it is also necessary that any given point of the trajectory is reached by the estimate at the same time instant. STOCHASTIC STATE ESTIMATION which is the state estimate propagation equation of the Kalman filter. that is. Matlab routines available through the class Web page implement a Kalman filter (with naive numerics) to estimate the state of that system from simulated observations. Figure 7. by the update equations ˆk|k x −1 Pk|k and P0|−1 = P0 . The error covariance matrix is easily propagated thanks to the linearity of the expectation operator: Pk+1|k = = = = E[(ˆk+1|k − xk+1 )(ˆk+1|k − xk+1 )T ] x x E[(Fk (ˆk|k − xk ) − ηk )(Fk (ˆk|k − xk ) − ηk )T ] x x T T Fk E[(ˆk|k − xk )(ˆk|k − xk )T ]Fk + E[ηk ηk ] x x T Fk Pk|k Fk + Qk where the system noise ηk and the previous estimation error ˆk|k − xk were assumed to be uncorrelated.6 Results of the Mortar Shell Experiment In section 7.5 shows the 2-norm of the covariance matrix over time. ˆ0|−1 = ˆ0 x x both assumed to be given.3 Kalman Filter Equations ∆ ∆ In summary.11) can be expanded as follows: yk = = = Hk xk + ξk = Hk (Fk−1 xk−1 + Gk−1 uk−1 + ηk−1 ) + ξk Hk Fk−1 xk−1 + Hk Gk−1 uk−1 + Hk ηk−1 + ξk Hk Fk−1 (Fk−2 xk−2 + Gk−2 uk−2 + ηk−2 ) + Hk Gk−1 uk−1 + Hk ηk−1 + ξk . the dynamic system equations for a mortar shell were set up. 7.4 shows that the distance between estimated and true target position does indeed converge to zero. Figure 7.5. obtaining ˆ0|k . amounts to solving a linear x system of equations with suitable weights for its rows.96 CHAPTER 7. and this occurs in time for the shell to be shot down. Notice that coincidence of the trajectories does not imply that the state estimate is up-to-date. = ˆk|k−1 + Kk (yk − Hk ˆk|k−1 ) x x −1 T −1 = Pk|k−1 + Hk Rk Hk where the Kalman gain is defined as and by the propagation equations T −1 Kk = Pk|k Hk Rk ˆk+1|k x Pk+1|k = = Fk ˆk|k + Gk uk x T Fk Pk|k Fk + Qk .7 Linear Systems and the Kalman Filter In order to connect the theory of state estimation with what we have learned so far about linear systems.2. Figure 7.3 shows the true and estimated trajectories. 7. the Kalman filter evolves an initial estimate and an initial error covariance matrix.

time 2 1.4: The estimate actually closes in towards the target.6 1.2 0 0 5 10 15 20 25 30 Figure 7. distance between true and estimated missile position vs.2 0 −0.2 1 0.8 0.4 0.7. .6 0.2 1 0.3: The true and estimated trajectories get closer to one another.7. Trajectories start on the right.4 1.6 0.8 1.8 0.4 1. LINEAR SYSTEMS AND THE KALMAN FILTER 97 true (dashed) and estimated (solid) missile trajectory 1.2 5 10 15 20 25 30 35 Figure 7.4 0.

22) where Φ(l.22) for every value of k = 0. The key thing to notice about this somewhat intimidating expression is that for any k it is a linear system in x0 . the initial state of the system. F1 η0 + . . j)Gj−1 uj−1 + νk (7. . . . = Hk Fk−1 Fk−2 xk−2 + Hk (Fk−1 Gk−2 uk−2 + Gk−1 uk−1 ) + Hk (Fk−1 ηk−2 + ηk−1 ) + ξk Hk Fk−1 . F1 G0 u0 + . . downwards ones to state update. .98 40 CHAPTER 7. + Gk−1 uk−1 ) + Hk (Fk−1 . We can write one system like the one in equation (7. = . yK (7. .23) . STOCHASTIC STATE ESTIMATION norm of the state covariance matrix vs time 35 30 25 20 15 10 5 0 0 5 10 15 20 25 30 Figure 7. j) = and the term k Fl . .  . k yk = Hk Φ(k − 1. . . . F0 x0 + Hk (Fk−1 . .   . Fj 1 for l ≥ j for l < j νk = Hk j=1 Φ(k − 1. and we obtain a large system of the form zK = ΨK x0 + gK + nK where  zK =  y0  . Upwards segments correspond to state propagation. . 0)x0 + Hk j=1 Φ(k − 1. K. . . .5: After an initial increase in uncertainty. j)ηj−1 + ξk is noise. + ηk−1 ) + ξk or in a more compact form. where K is the last time instant considered. . the norm of the state covariance matrix converges to zero. .

23) all at once. In some applications. 0) 0 H1 G0 u0 . under the assumption that the noise that affects each of the equations has the same probability distribution. We know that if ΨK has full rank. that is. and not only for the estimate ˆ0|k of the initial state. . .23 are not equally important. . to obtain an estimate of x0 from the measurements y0 . so we ought to be able to do better. The system (7. In summary. so that Ψ† = (ΨT ΨK )−1 ΨT . However. the Kalman filter differs from a linear solver in the following important respects: 1. the best we can do is to solve the system zK = ΨK x0 + gK in the sense of least squares. and the Kalman filter makes optimal use of this information for a proper weighting of each of the scalar equations in (7. 2.23). a small covariance means that we believe in that measurement.23 are equally important.7. A solution for the estimate ˆk|k of the current state is given. . having better and better approximations to the solution as new data come in is much preferable in a dynamic setting. an initial solution is refined over time as new measurements become available. . . The quantitative embodiment of this intuitive idea is at the core of the Kalman filter. x x 3. K)GK−1 uK−1 )  ν0  . knowledge of the initial state may obsolesce and become less and less useful. LINEAR SYSTEMS AND THE KALMAN FILTER  ΨK   =    gK   =    H0 H1 F0 .           99 nK HK (Φ(K − 1. 1)G0 u0 + . This is equivalent to assuming that all the noise terms in nK are equally important. The final solution can be proven to be exactly equal to solving system (7. . that all the noise terms in nK in equation 7. The Kalman filter computes up-to-date information about the current state. + Φ(K − 1. the Kalman filter for a linear system has been shown to be equivalent to a linear equation solver. As time goes by. Intuitively. Rather. K K K The least square solution to system (7. However. the result with the pseudoinverse is the K same as we would obtain by solving the normal equations. we know the covariance matrices of all these noise terms. and this is in fact the case. However. Measurements come with covariance matrices. νK Without knowing anything about the statistics of the noise vector nK in equation (7. .23). and therefore in that equation. and weigh each equation to keep these covariances into account.  =  . yK : ˆ0|K = Ψ† (zK − gK ) x K where Ψ† is the pseudoinverse of ΨK . The noise terms in nK in equation 7. . which should consequently be weighed more heavily than others.  . . HK Φ(K − 1. data my never stop arriving. Better information ought to yield more accurate results. . .23) minimizes the residue between the left and the right-hand side under the assumption that all equations are to be treated the same way. where one cannot in general wait for all the data to be collected.23) is not solved all at once.7.

Sign up to vote on this title
UsefulNot useful