Elements of Mathematical Style

A work in progress

Bruce Francis

Revised July 19, 2010

downloadable at http://sites.google.com/site/brucefranciscontact/
Thanks

I thank Murray Wonham for reading this work and offering helpful comments.

1
Contents

1 Introduction 3

2 Sets and Functions 4

3 Elements of Logic 6

4 Definition and Theorem Statements 13

5 The Elements of Style 18

6 Giving a Talk 22

2
Chapter 1

Introduction

This is a very brief handbook on writing documents with mathematical content: reports, theses,
journal papers, and conference papers. The handbook is written primarily for graduate students in
control and similar subjects.
The handbook begins with elementary mathematics. To write your theorems correctly, you need
to know elementary logic, namely, propositional calculus. To understand this logic, you need to
know about sets and functions. The next chapter gives a brief review of sets, functions, and the
relevant notation. The following chapter gives a brief presentation of the logic required to state
theorems. The key point is that variables (ε, δ, etc.) must be quantified (“there exists,” “for all”)
and the quantifiers must be in the right order. With this background, we can proceed to theorem
statements.
Following this technical part is presented a style guide, à la Strunk and White, for mathematical
writing, and then for giving a talk using computer slides.

3
Chapter 2

Sets and Functions

Examples of sets: the set R of real numbers; the set Rm×n of m × n real matrices; the set F(U, V )
of functions from the set U to the set V . A set is frequently defined using notation like this:

S = {x : x ∈ R, x > 0}.

In words this says, S is the set of all x such that x belongs to the real numbers and x is positive.
Thus a colon means “such that,” a comma means “and,” and ∈ means “belongs to” or “is a member
of.” Another common form of writing the set is to use a vertical line instead of a colon:

S = {x|x ∈ R, x > 0}.

Suppose you want the set of numbers greater than a number a. This is right:

S = {x : x ∈ R, x > a}, a ∈ R.

This is wrong:

S = {x : x ∈ R, x > a, a ∈ R}.

In fact, this latter set is all of R, for it says S is the set of all x such that x belongs to the real
numbers, x is greater than a, and a belongs to the set of real numbers too, i.e., a is not fixed.
If U and V are two sets, a function from U to V is a rule that assigns to every element of U
an unambiguous element of V —there cannot be two possible different values f (u) for the same u
in U . The terms function, mapping, and transformation are synonymous. The notation

f : U −→ V

means that f is a function from U to V . The set U is called the domain of f and V is called the
codomain. It is sometimes useful to avoid giving a name to the function, and then we use the
“mapsto” symbol 7→. For example, the function x 7→ x2 from R to itself takes a real number x and
squares it. This function could alternatively be written

f : R −→ R, f (x) = x2 .

Notice that the symbol for the function is f whereas, strictly speaking, f (x) denotes the value of f
at the point x. That being understood, it is common to say “the function f (x).”

4
CHAPTER 2. SETS AND FUNCTIONS 5

Example
This example is in the context of elementary calculus, or signal analysis. Let N+ denote the set of
positive integers 1, 2, 3, . . . . Consider a sequence x[1], x[2], . . . of real numbers. For example, x[1]
could be the first voltage reading of a circuit element, x[2] the second voltage reading, and so on.
This sequence defines a function

x : N+ −→ R,

namely, x maps an integer n to the voltage value x[n]. Thus x is an element of the set F(N+ , R) of
all functions from N+ to R.
Some sequences, like x[n] = 1/2n , converge and others, like x[n] = (−1)n , do not. Let {T, F }
be the set of two truth values, true (T) and false (F), and define the function

T if x[n] converges
P (x) =
F if x[n] does not converge.

Then P is a function F(N+ , R) −→ {T, F }. The value P (x) is true if and only if x[n] is a convergent
sequence.

Example
Suppose A is an n × n real matrix, where n is fixed. That is, A ∈ Rn×n . Then the determinant of
A, det(A), is a function Rn×n −→ R.
Let C denote the set of complex numbers. The set σ(A) of eigenvalues of A is a function too;
its domain is Rn×n and its codomain could be taken as the set of all subsets of C, which is denoted
by 2C and which is called the power set of C.
Define

T if det(A) 6= 0
P (A) =
F if det(A) = 0.

Then P is a function Rn×n −→ {T, F }. The value P (A) is true if and only if A is invertible.
Chapter 3

Elements of Logic

It will be convenient to use the abbreviation iff for “if and only if.” As an example use, let A be
a square real matrix and consider the homogeneous equation Ax = 0, where x and 0 are vectors.
The vector x = 0 is always a solution of the equation, but there may be other solutions. Here’s a
result about uniqueness:
The equation Ax = 0 has the unique solution x = 0 iff the determinant of A is nonzero.
This statement contains two logical implications, the “if” part and the “only if” part. The “if” part
is this: The equation has a unique solution if the determinant of A is nonzero. That is, det(A) 6= 0
is a sufficient condition for there to be a unique solution. The “only if” part is this: The equation
has a unique solution only if the determinant of A is nonzero. That is, if det(A) = 0, then there is
not a unique solution.
Let R denote the set of real numbers. On that set consider the statement “x > 0.” For every x
in R, either “x > 0” is true (T) or it is false (F). This permits us to define a function
P : R −→ {T, F }
by saying P (x) = T if x > 0 and P (x) = F if x ≤ 0. We say that P is a proposition defined on R;
P (x) is the proposition that x is positive. Generalizing, let S be any set. A proposition on S is a
function
P : S −→ {T, F }.
The negation of P is the proposition, denoted by ¬P , that is defined by

T if P (x) = F
¬P (x) =
F if P (x) = T.

Example
On the set R of real numbers, consider the statement “1 ≤ x ≤ 10” that x is greater than or equal
to 1 and less than or equal to 10. This defines a proposition P , namely, P (x) = T iff 1 ≤ x ≤ 10.
In words, P (x) is the proposition that x is between 1 and 10, inclusive. The negation ¬P is the
proposition of being either less than 1 or greater than 10.

6
CHAPTER 3. ELEMENTS OF LOGIC 7

Example
Let S = F(R, R), the set of all functions R −→ R, and consider the statement “|f (x)| < 1” about f
in S and x in R. This statement defines a proposition P on the set S × R by saying P (f, x) = T iff
|f (x)| < 1. On the other hand, suppose f is given, i.e., fixed and therefore not a variable. Then we
could regard “|f (x)| < 1” as a proposition of x alone. Similarly, the statement “|f (x)| < ε” could
define a proposition P (f, x, ε) if all three arguments are variables, or P (x, ε) if f is given, or even
P (ε) if both f and x are given.

There are ways to combine two propositions. Let P and Q be propositions on the same set S.
Their conjunction, written P ∧ Q, is the proposition that is true iff both P and Q are true. It is
convenient sometimes to use a comma and to write P, Q instead of P ∧ Q. For example, on R, if P
is the proposition “x ≤ 2” and Q the proposition “x > 0,” then their conjunction is the proposition
“x ≤ 2, x > 0,” that is, “0 < x ≤ 2.” The disjunction of P and Q, written P ∨ Q, is true iff one
or both are true. A truth table is a convenient way to define conjunction and disjunction:
P ∧Q P Q P ∨Q P Q
T T T T T T
F T F T T F
F F T T F T
F F F F F F
Two propositions P and Q defined on the same set are said to be equivalent, written P ≡ Q,
if the two functions are equal, i.e., P (x) = Q(x) for every x.
The following negations can be verified:
¬(P ∧ Q) ≡ (¬P ) ∨ (¬Q)
¬(P ∨ Q) ≡ (¬P ) ∧ (¬Q).

Example
Let S = F(R, R). Let P be the proposition of being piecewise continuous, that is, P (f ) = T iff
f is piecewise continuous, and let Q be the proposition of being exponentially bounded. Then the
negation of “f is piecewise continuous and exponentially bounded,” that is, P (f ) ∧ Q(f ), is “either
f is not piecewise continuous or it is not exponentially bounded.”

There is another proposition derived from two given propositions. Let P and Q be two propo-
sitions on the same set. Define a new proposition, called implication, denoted by P ⇒ Q:
P ⇒Q P Q
T T T
F T F
T F T
T F F
CHAPTER 3. ELEMENTS OF LOGIC 8

Thus P ⇒ Q is true for all truth values of P and Q except when P is true and Q is false. Therefore

(P ⇒ Q) ≡ ((¬P ) ∨ Q). (3.1)

In view of (3.1), the negation of implication satisfies

¬(P ⇒ Q) ≡ (P ∧ ¬Q).

Example
Consider the set R. Let P be the proposition “1 < x < 6,” let Q be the proposition “−5 < x < 5,”
and let R be the proposition P ⇒ Q. In view of (3.1), R is the proposition

x ≤ 1 or x ≥ 6 or − 5 < x < 5,

equivalently, x does not satisfy 5 ≤ x < 6. That is, R(x) = T for all x except when 5 ≤ x < 6. In
this latter interval, P is true but Q is false.

Let P and Q be propositions and let R denote the proposition P ⇒ Q. The following rule of
inference is obvious: For a given x,

if R(x) = T and P (x) = T , then Q(x) = T .

This can be verified from the truth table. The rule says the obvious thing: If the implication P ⇒ Q
is true and if P is true, then Q is true. Thus if R is true everywhere (on the set on which it is
defined), then if P is true, so too is Q; in this case we say that P is a sufficient condition for Q,
equivalently, Q is a necessary condition for P .

Example
Consider the 2nd order polynomial p(s) = s2 +a1 s+a0 , with real coefficients. Let a denote the vector
of coefficients (a0 , a1 ), let P (a) be the proposition that the roots of p(s) have negative real parts,
let Q(a) be the proposition that the coefficients ai are both positive, and let R be the proposition
P ⇒ Q. You can easily check that R(a) is true for every a. Thus, if P is true, so is Q. We can also
say that Q is necessary for P , i.e., a necessary condition for all roots to be in the left half-plane is
that the coefficients be positive. In fact, the converse direction Q ⇒ P is always true too. It follows
that P and Q are equivalent: P ≡ Q. We say P is true iff Q is true.
Extend the discussion to the 3rd order polynomial p(s) = s3 + a2 s2 + a1 s + a0 . Again, let a
denote the vector of coefficients, let P (a) be the proposition that the roots of p(s) have negative
real parts, let Q(a) be the proposition that the coefficients ai are all positive, and let R be the
proposition P ⇒ Q. It is a theorem that R(a) is true for every a. Thus, if P is true, so is Q. The
converse, however, is false. That is, there exist coefficients such that Q(a) is true but P (a) is false.
CHAPTER 3. ELEMENTS OF LOGIC 9

We frequently want to say that a certain proposition is true everywhere on the set on which it
is defined. If S is the set and P the proposition, this is written

(∀x ∈ S) P (x) = T.

In words: For every x in S, P (x) is true. The symbol ∀ (upside down A) is called the universal
quantifier and it means “for all” or “for every.” Notice that the quantification, ∀x ∈ S, is given in
brackets before the proposition is stated. If S is understood, we could write

(∀x) P (x) = T.

Or to abbreviate even further we could write

(∀x) P.

There is a second quantifier, the existential quantifier (backwards E):

(∃x ∈ S) P (x) = T.

This means, there exists an x in S such that P (x) is true. Notice that the words “such that” are
not written but are understood. Again, we might abbreviate to

(∃x) P (x) = T

or to

(∃x) P.

The negation of (∀x ∈ S) P (x) is written ¬(∀x ∈ S) P (x), and you can check that

¬(∀x ∈ S) P ≡ (∃x ∈ S) ¬P.

The meaning for either side is that P is false for at least one element of S. Here are some other
negations, in simplified form:

¬(∃x) P ≡ (∀x) ¬P

¬((∀x) P ∧ Q) ≡ (∃x) ((¬P ) ∨ (¬Q))

¬((∀x) P ∨ Q) ≡ (∃x) ((¬P ) ∧ (¬Q)).
Consider negating the statement (∀x)(∃y) P (x, y). The statement has the form (∀x) Q(x), where
Q(x) is the proposition (∃y) P (x, y). It follows that

¬(∀x)(∃y) P (x, y) ≡ (∃x)¬(∃y) P (x, y) ≡ (∃x)(∀y) ¬P (x, y).

CHAPTER 3. ELEMENTS OF LOGIC 10

Example
Consider a sequence {an }n≥1 of real numbers. We are going to define what it means for the sequence
to be bounded. But the sequence itself is fixed in what follows. Here is the definition:

(∃B ≥ 0)(∀n ≥ 1) |an | ≤ B. (3.2)

This statement is again parsed from left to right. In words, (3.2) says this: There exists a non-
negative number B such that, for every positive integer n, the absolute value of an is bounded by
B. Notice in (3.2) that the two quantifier phrases, ∃B ≥ 0 and ∀n ≥ 1, are placed in brackets and
precede the term |an | ≤ B. This latter term has n and B as variables in it that need to be quanti-
fied. We cannot say merely that {an }n≥1 is bounded if |an | ≤ B (unless it is known or understood
what the quantifiers on n and B are). In general, all variables in a statement need to be quantified.
Notice that (3.2) has the form

(∃B)(∀n) P (B, n).

The negation of this is

(∀B)(∃n) ¬P (B, n).

Therefore, the negation of (3.2) is

(∀B ≥ 0)(∃n ≥ 1) |an | > B.

The order of quantifiers is crucial. Observe that (3.2) is very different from saying

(∀n ≥ 1)(∃B ≥ 0) |an | ≤ B, (3.3)

which is true of every sequence, not just bounded ones..

Example
Here’s the definition that the sequence {an }n≥1 converges to 0:

(∀ε > 0)(∃N ≥ 1)(∀n ≥ 1) n ≥ N ⇒ |an | < ε. (3.4)

This says, for every positive ε there exists a positive N such that, for every positive n, if n ≥ N
then |an | is less than ε. In words: an is arbitrarily small if n is sufficiently large. The structure of
(3.4) is

(∀ε)(∃N )(∀n) P (N, n) ⇒ Q(n, ε).

The negation is therefore

(∃ε)(∀N )(∃n) P (N, n) ∧ ¬Q(n, ε),

CHAPTER 3. ELEMENTS OF LOGIC 11

that is,
(∃ε > 0)(∀N ≥ 1)(∃n ≥ 1) n ≥ N, |an | ≥ ε.
Statement (3.4) is frequently contracted to
(∀ε > 0)(∃N ≥ 1)(∀n ≥ N ) |an | < ε.
Some people write (3.4) in this way,
(∀ε > 0)(∃N (ε) ≥ 1)(∀n ≥ 1) n ≥ N (ε) ⇒ |an | < ε,
to emphasize that N may depend on ε. This is entirely unnecessary, since the left-to-right order
already allows N to depend on ε.

Example
The function f : R −→ R is continuous at the point x = a iff
(∀ε > 0)(∃δ > 0)(∀x) |x − a| < δ ⇒ |f (x) − f (a)| < ε. (3.5)
The negation is therefore
(∃ε > 0)(∀δ > 0)(∃x) |x − a| < δ, |f (x) − f (a)| ≥ ε.

Example
The following statement is a theorem in Complex Variables and Applications, by Churchill and
Brown:
If a function is analytic at a point, then its derivatives of all orders are also analytic at
that point.
Let us write the logical negation of the statement. We may do so by writing the statement in logic
form. The variables in the statement are the function, say f , the point, say a, and the order, say
m. Then “f is analytic at a” defines a proposition P (f, a), and “the mth -derivative of f is analytic
at a” defines a proposition Q(f, m, a). Thus the theorem statement has the form
(∀f )(∀a) P (f, a) ⇒ (∀m) Q(f, m, a).
The negation is therefore
(∃f )(∃a) P (f, a), (∃m) ¬Q(f, m, a).
In words: There is some function and some point such that the function is analytic at the point but
some derivative of the function is not analytic at the point.

Finally, we illustrate the logic structure of a proof by an example.

CHAPTER 3. ELEMENTS OF LOGIC 12

Example
To be proved The step function u(t), that equals 0 for t < 0 and 1 for t ≥ 0, is not continuous at
t = 0.

Proof The definition that a function f (t) is continuous at t = 0 is this:

(∀ε > 0)(∃δ > 0)(∀t)|t| < δ ⇒ |f (t) − f (0)| < ε.

Thus we are to prove

(∃ε > 0)(∀δ > 0)(∃t) |t| < δ, |u(t) − u(0)| ≥ ε.

We proceed from left to right, being as specific as possible.

1. Let ε = 1/2.

2. Let δ > 0 be arbitrary.

3. Let t = δ/2.

4. Then |t| < δ and |u(t) − u(0)| = 1 > ε.

Chapter 4

Definition and Theorem Statements

Definition Statements
A mathematical definition has to be grammatically correct according to the grammar of mathemat-
ics, that is, the rules of logic. It usually isn’t written in logic symbols, but it has to have the same
form. So, all variables have to be quantified.

Example
This example discusses the definition of bounded function. The first attempt is incorrect because
it leaves out the quantifiers:

A real-valued function x(t) of the real variable t is bounded if |x(t)| ≤ B.

The next attempt has quantifiers but in the wrong order and therefore it is incorrect:

A real-valued function x(t) of the real variable t is bounded if for all t there is a real
B such that |x(t)| ≤ B.

The next attempt has quantifiers but not in the best places:

A real-valued function x(t) of the real variable t is bounded if there is a real B such
that |x(t)| ≤ B for all t.

The next one is best:

A real-valued function x(t) of the real variable t is bounded if there exists a real number
B such that, for all t, |x(t)| ≤ B.

Notice that the last one has, as it should, the same structure as the logic statement that x(t) is
bounded:

(∃B)(∀t) P (B, t).

13
CHAPTER 4. DEFINITION AND THEOREM STATEMENTS 14

Example
This example concerns stability theory. Consider the differential equation ẋ = f (x), where x is a
vector-valued function of time t. Assume x = 0 is an equilibrium point, i.e., f (0) = 0. Here’s the
definition in logic format that the origin is a stable equilibrium point:

(∀ε > 0)(∃δ > 0)(∀x(0)) kx(0)k < δ ⇒ (∀t ≥ 0) kx(t)k < ε.

Here’s the same thing in words:

For every ε > 0 there exists δ > 0 such that if the state starts in the ball of radius δ, it
will remain forever in the ball of radius ε.

Here’s a simplified verbal version:

The solution will remain arbitrarily close to the origin if it starts sufficiently close to the
origin.

Example
Here’s a definition of the concept of safety:

Given a set of admissible states X ⊂ Rn , a set of initial states X0 ⊂ X , a set of unsafe

states Xu ⊂ X , and a set of admissible inputs U ⊂ Rm , the safety property for

ẋ = f (x, u), x ∈ Rn , u ∈ U

holds if there exists no T ≥ 0 with input trajectory u : [0, T ] −→ U such that x(0) ∈ X0 ,
x(T ) ∈ Xu , x(t) ∈ X for all t ∈ [0, T ].

The idea in this definition is that an unsafe situation has occurred if x(t) has entered the unsafe
set Xu at some time; hence, the system is unsafe if x(t) can enter Xu for some x(0) and some u. Of
course, the case x(0) ∈ Xu must be ruled out, and that is why the set X0 is needed. Similarly, there
may be control values we may wish to rule out, so bring in a set U of permitted control values. The
reason for X is unclear, so let’s leave that set out. This leads us to the tentative definition that the
system is unsafe if

(∃x(0))(∃T > 0)(∃u) (∀t, 0 ≤ t ≤ T ) u(t) ∈ U, x(0) ∈ X0 , x(T ) ∈ Xu .

Of course, it is understood that u must be as stated before: bounded and piecewise continuous.
Are any other quantifiers required? The definition is about the system and the sets X0 , Xu , U, so
these should not be quantified. Also, x(T ) is uniquely determined by x(0), T, u, so it should not be
quantified. With these thoughts, here’s a proposed correction:

Consider the system

ẋ = f (x, u)
CHAPTER 4. DEFINITION AND THEOREM STATEMENTS 15

where f maps Rn ×Rm to Rn . Assume the input u is bounded and piecewise continuous.
Assume also that f is sufficiently regular that solutions x(t) exist for all x(0) and all such
inputs. Let X0 ⊂ Rn be a set of allowed initial states, Xu ⊂ Rn a set of states deemed to
be unsafe, and U ⊂ Rm a set of admissible input values. The system is unsafe if there
exist x(0), T > 0, and u, with x(0) in X0 and u(t) in U for all 0 ≤ t ≤ T , such that x(T )
is unsafe, i.e., x(T ) ∈ Xu . The system is safe if it is not unsafe, i.e., such x(0), T , u do
not exist.

Theorem Statements
Theorem statements are similar: They have to be grammatically correct according to the grammar
of mathematics.

Example
Theorem Consider system (1). Assume (A, B) is controllable and (C, A) is observable.
Let x(0) be arbitrary. There exists an LTI controller, with input yref − y and output u,
such that the feedback system is stable and the tracking error goes to zero as t goes to
infinity.

Let us analyze this statement by putting it in logic notation. We have to quantify all variables.
What are the variables? Let’s say that system (1) is a number of state-space equations, parametrized
by matrices A, B, . . . and certain types of inputs. Is “system (1)” a variable? The sensible thing
is to fix the structure of the equations but not the actual matrices. So our logic statement would
start like this:
(∀A, B, C, . . . )(A, B) controllable and (C, A) observable ⇒ .
After this there will be the term (∃ LTI controller) and since this term comes after the quantification
on A, B, C, . . . the implication is that the controller depends on A, B, C, . . . , which is natural. What
about the phrase “Let x(0) be arbitrary.” There’s only one way to quantify that, (∀x(0)). So it
seems the theorem is of this form:
(∀A, B, C, . . . )(A, B) controllable and (C, A) observable ⇒ (∀x(0))(∃ LTI controller) . . . .
But this says the controller depends on x(0), that is, if the initial plant state changes, the controller
has to be redesigned. This must be wrong. The correct form is
(∀A, B, C, . . . )(A, B) controllable and (C, A) observable ⇒ (∃ LTI controller)(∀x(0)) . . .
In this case, the theorem statement should have been
Theorem Consider system (1). Assume (A, B) is controllable and (C, A) is observable.
There exists an LTI controller, with input yref − y and output u, such that the feedback
system is stable and, for every x(0), the tracking error goes to zero as t goes to infinity.
CHAPTER 4. DEFINITION AND THEOREM STATEMENTS 16

Example
This example concerns the Nyquist criterion:

A feedback system is stable if and only if the contour ΓL in the L(s)-plane does not
encircle the (−1, 0) point when the number of poles of L(s) in the right-hand s-plane is
zero.

The phrase “when the number of poles ...” seems to be a quantifier of the form

(∀L) the number of poles of L(s) in the right-hand s-plane is zero . . .

But this cannot be. The statement concerns a given feedback system with open-loop transfer
function L(s). Thus, L(s) is not a variable and therefore can’t be quantified. Then what does the
subordinate clause “when . . . ” mean? In fact the clause should be replaced by an assumption. The
next mistake concerns the ambiguity of the verb “encircle.” What if the contour passes through
the point (−1, 0)? Does that count as not encircling it? The correct statement (though not the full
Nyquist criterion) is this:

Assume the open-loop transfer function L(s) is stable. Then the feedback system is stable
if and only if the contour ΓL does not pass through or encircle the point (−1, 0).

Example
The next example concerns the converse Lyapunov theorem.

Theorem Let x = 0 be an equilibrium point for the nonlinear system

ẋ = f (t, x)

where f : [0, ∞) × D −→ Rn is continuously differentiable, D = {x ∈ Rn : kxk < r}, and

the Jacobian matrix ∂f /∂x is bounded on D, uniformly in t. Let k, γ and r0 be positive
constants with r0 < r/k. Let D0 = {x ∈ Rn : kxk < r0 }. Assume that the trajectories of
the system satisfy

kx(t)k ≤ kkx(t0 )ke−γ(t−t0 ) , ∀x(t0 ) ∈ D0 , ∀t ≥ t0 ≥ 0

Then there is a function V : [0, ∞) × D −→ R that satisfies the inequalities

c1 kxk2 ≤ V (t, x) ≤ c2 kxk2

∂V ∂V
+ f (t, x) ≤ −c3 kxk2
∂t ∂x

∂V
∂x ≤ c4 kxk

for some positive constants c1 , c2 , c3 , and c4 .

CHAPTER 4. DEFINITION AND THEOREM STATEMENTS 17

This theorem statement is self-contained, and therefore includes the setup, assumptions, notation—
everything. There are a few things missing, such as assumptions about the differentiability of V and
a quantifier on t in the latter three inequalities. The quantifiers that are given are not uniformly
placed: “there is a function V ” precedes the inequalities while “for some positive constants” follows
them. Is t0 quantified or not? It should be. Finally, by saying the system is nonlinear, the theorem
may seem to be ruling out linear systems. But of course the theorem applies to the case where
f (t, x) is linear in x.
Many people like theorem statements to be self-contained, with all the definitions, notation,
etc. Then they don’t have to hunt for these things. But the danger is that you may not see what
the main point of the theorem is. The above theorem’s main point is that, if the origin is locally
exponentially stable, then there exists a Lyapunov function. Here’s an attempt to bring that out
by introducing definitions before the theorem statement:
The theorem to follow concerns the system

ẋ = f (t, x), (4.1)

where x(t) is a vector in Rn . Three preliminary assumptions: The equation has a

unique solution for every initial time t0 ≥ 0 and every initial state x(t0 ), and the solution
exists for all t > t0 ; the origin is an equilibrium, that is, for all t ≥ 0, f (t, 0) = 0; the
Jacobian matrix ∂f /∂x is bounded for sufficiently small kxk, uniformly in t. Next, some
notation: For a positive number r, let B(r) denote the open ball in Rn of radius r.
The theorem concerns two key concepts: local exponential stability and local Lyapunov
function. Let us define those concepts. The origin is locally exponentially stable
if, for every t0 , there exist positive k, γ, r such that, for every x(t0 ) ∈ D(r) and every
t ≥ t0 ,

kx(t)k ≤ kkx(t0 )ke−γ(t−t0 ) .

A Lyapunov function for system (4.1) is a continuously differentiable function V :

[0, ∞) × Rn −→ R having the property that there exist positive c1 , c2 , c3 such that for
all t ≥ 0 and x ∈ Rn

c1 kxk2 ≤ V (t, x) ≤ c2 kxk2

∂V ∂V
+ f (t, x) ≤ −c3 kxk2
∂t ∂x

∂V
∂x ≤ c4 kxk.

A local Lyapunov function is defined similarly except that V (t, x) is defined only for
kxk sufficiently small, and then the three inequalities are required to hold only for such
x.
Theorem Consider system (4.1) along with the three preliminary assumptions. If the
origin is locally exponentially stable, then there exists a local Lyapunov function.
Chapter 5

The Elements of Style

P.R. Halmos is a mathematician renowned for his writing. It is simple, very clear, seemingly
informal, but perfectly rigorous. Let us look at an excerpt of the 1948 edition of his book Finite
Dimensional Vector Spaces. I have modernized the punctuation slightly:
Before studying linear functionals and conjugate spaces in more detail we wish to in-
troduce a notation that may appear weird at first sight but which will clarify many
situations later on. We have already used the trick of denoting such a composite object
as a linear functional, y(x), by a single letter y. Sometimes, however, it is necessary to
use the function notation and indicate somehow that if y is a linear functional on X and
if x is a vector in X , then y(x) is a fixed scalar. The notation we propose to adopt here,
instead of writing y followed by x in parentheses, is x and y enclosed between square
brackets and separated by a comma: In other words we shall write [x, y] in place of
y(x). Because of the unusual nature of this notation we shall expend on it some further
verbiage.
As we have just pointed out, [x, y] is a substitute for the function notation y(x); [x, y] is
the scalar value we obtain if we take the value of the linear functional y at the vector x.
Let us take an analogous situation (but not concerned with linear functionals). Let x be
a real variable; let y = y(x) be the function y(x) = x2 . The notation [x, y] is a symbolic
way of writing down the recipe for the actual operations performed; it corresponds to
the sentence [take a number x, and square it].
Using this notation we may sum up: To every vector space X we make correspond the
conjugate space X 0 consisting of all linear functionals of X ; to every pair x and y, where
x is a vector in X and y is a linear functional in X 0 , we make correspond the scalar [x, y]
defined to be the value of y at x. In terms of the symbol [x, y] the defining property of
a linear functional is

[α1 x1 + α2 x2 , y] = α1 [x, y1 ] + α2 [x2 , y],

and the definition of the linear operations on linear functionals is

[x, α1 y1 + α2 y2 ] = α1 [x, y1 ] + α2 [x, y2 ].

The two relations together are expressed by saying that [x, y] is a bilinear functional
of the vectors x in X and y in X 0 .

18
CHAPTER 5. THE ELEMENTS OF STYLE 19

Notice that Halmos has left out many quantifiers. He is able to do so because his writing is so clear
that the reader can easily insert them. That is, the quantifiers are understood in his writing. The
statement

[α1 x1 + α2 x2 , y] = α1 [x, y1 ] + α2 [x2 , y]

is understood to mean

(∀x1 , x2 , y, α1 , α2 ) [α1 x1 + α2 x2 , y] = α1 [x, y1 ] + α2 [x2 , y].

For new writers it is advisable to put the quantifiers in.

With that as an inspiring example, we turn to some rules of style.

1. Choose notation carefully and be consistent.

Matrices are usually denoted by capital Roman letters; vectors by lower case Roman. There’s
no need to write a vector as ~x unless it’s the usual convention, as in physics. Don’t change
alphabets for no reason: x(t) and ω(t) shouldn’t be the same sort of object. Otherwise, the
reader has a harder time because the meanings of symbols are not transparent; the reader
loses confidence in the writer. In general, your notation should be simple and intuitive.

2. Write your definitions in causal format.

Here’s a noncausal definition:

Let {Ln } be a sequence, where Ln belongs to Hom(X , Y), where Hom stands for
homomorphism, and where X and Y are vector spaces.

This is annoying because the reader doesn’t fully comprehend until the end of the sentence.
Here’s the causal form, in which the reader understands from the beginning:

Let X and Y be vector spaces and let Hom(X , Y) denote the space of homomor-
phisms from X to Y. For each integer n, let Ln belong to Hom(X , Y).

3. Have a consistent punctuation for mathematical phrases.

Treat each mathematical phrase as though it were a noun. Example:

In general Ax = b is solvable iff b belongs to the column span of A.

In this sentence “Ax = b” is treated as a noun; the equal sign is not treated as a verb. Here’s
some text with a displayed mathematical phrase:

The main result on the solvability of linear equations is

 
(∃x) Ax = b ⇐⇒ rank A b = rank A.

The displayed statement is treated as a noun that ends the sentence, hence the period. Two
consecutive equations are treated as one noun; so there is a period only at the end of the last
one and no comma after the first one. Example:
CHAPTER 5. THE ELEMENTS OF STYLE 20

We consider the usual state model:

ẋ = Ax + Bu
y = Cx + Du.

4. Avoid dangling participles.

It is so convenient in mathematical writing to use this construction: Substituting (3) into (1),
the state equation becomes . . . . But it is grammatically wrong. Who is doing the substituting?
Presumably the writer or the reader. But as the sentence is written, the only noun that could
be doing the substituting is “state equation,” and that’s not what the writer wants to say.
There are two possible fixes: Make “substituting” a noun: Substituting (3) into (1) results
in the state equation . . . ; or put in the proper noun doing the substituting: Substituting (3)
into (1), we arrive at the state equation . . . .

5. Give the example before the theorem, not after.

In general it is better to go from the specific to the general. It should be easier to understand
an example than a formal theorem statement. Indeed, a well-designed example is probably
more revealing than the full theorem statement.

6. Don’t put definitions or new notation in theorem statements.

A theorem statement should be as succinct as possible. Put the preamble before the theorem
statement so that the result itself stands out clearly.

7. Don’t start a sentence with a mathematical symbol.

It may conflict with the convention that sentences begin with capital letters.

8. Don’t use a colon where it doesn’t belong.

You wouldn’t write “My favourite movie is: The Bourne Identity.” So don’t write

The main result on the solvability of linear equations is:

 
(∃x) Ax = b ⇐⇒ rank A b = rank A.

But this is correct:

The main result on the solvability of linear equations is the following:

 
(∃x) Ax = b ⇐⇒ rank A b = rank A.

9. Use “that” before a restrictive clause and “which” before a nonrestrictive clause.
Examples:

The condition K 6= 0 is ruled out by our initial assumption, which we made only
because we were going to divide by K.
If a step, which is bounded, is applied to a stable system, the output is bounded
too.
If an input that is bounded is applied to a stable system, the output is bounded
too.
CHAPTER 5. THE ELEMENTS OF STYLE 21

10. The Latin equivalent of “that is” is “id est.” Its abbreviation therefore has a period after each
letter. It is always followed by a comma.
Example:

Assume x = 0 is an equilibrium point of the equation ẋ = f (x), i.e., f (0) = 0.

11. The Latin phrase “et alii” means “and others.” The phrase “et al.” is the abbreviation.
Notice that there is only one period.

12. The word “only” is a delimiter and should be placed immediately before that which it is
delimiting.
Examples:

She only walked to the store to buy milk. (Instead of also skipping or running, she
only walked.)
She walked only to the store to buy milk. (Instead of going also to the post office,
she went only to the store.)
She walked to the store only to buy milk. (Instead of also wanting to talk to the
owner or meet a friend, she went only to buy milk.)
She walked to the store to buy only milk. (Instead of also wanting to buy other
things.)
Wrong: A local Lyapunov function V (x) is only defined for kxk sufficiently small.
Right: A local Lyapunov function V (x) is defined only for kxk sufficiently small.
Chapter 6

Giving a Talk

Here we look at giving a talk using computer slides.

1. Coordinate the slides and what you’re saying.

You can’t prepare beautiful slides and then afterwards think about what you’re going to say.
Your slides should complement your speech.

2. This is very important: The audience cannot at the same time read text and listen to you
talk.
For this reason, it would be best to have no text on your slides: You would speak your
presentation and illustrate it with the slides. That is, on the slides would go diagrams,
pictures, movies, and mathematics—no text except for titles. Then the audience would not
be torn deciding whether to listen to you or read text.

3. If you do have text, you must read it.

If there is text, presumably you want the audience to read it. In order for that to happen,
you must either stop talking and let the audience read, or read it to them. If there is text
that you are not going to read, remove that text from your slides.

4. Remove from your slides everything that is unnecessary.

When you have finished preparing your talk, go over your slides and remove everything that
the audience does not need to see. For example, your university logo and banners. And while
you’re at it, do you really need background colour on the pages and animated transitions?
Aren’t these distractions from your message?

5. Say as early as possible in your talk what problem you’re going to present.
Get the audience interested in your problem as soon as possible. Don’t begin with boring
generalities, literature review, etc. Say what your problem is so that the audience can fix on
something specific. Do the generalities later—they won’t be so boring.

6. If your talk has theorems, do the examples first.

7. Use a pointer.

22
CHAPTER 6. GIVING A TALK 23

The audience must know at all times where you are on a slide. If your slide has several things
on it, you must point to what part you’re talking about. You can use the mouse/pointer, a
laser-pointer, or a real pointer. The latter two require you to turn away from the audience,
which is somewhat rude.

8. If your talk is at a conference, the purpose of your talk is to get people to read your paper in
the proceedings.
Thus, you don’t need to give all your results.