How to write mathematics

Effective mathematical communication is an important skill, yet is rarely taught. Effective

communication is important for three reasons:
1. It helps others (e.g. your peers, your teachers, your markers) to understand and ap-
preciate your ideas.

2. It enables you to understand your ideas, six months later, when you can’t remember
what you were thinking when you wrote them.

3. It enhances your ability to communicate in nonmathematical spheres:

• To clearly express scientific ideas.

• To write good computer code and clear documentation or specifications for such
code.
• To formulate clear and compelling arguments in any subject.

Effective mathematical communication is based on several principles.

Words vs. symbols

Mathematical prose relies heavily on symbols, but it is important not to use symbols to
excess. Effective communication requires complete, grammatically correct sentences. Sym-
bolism should never replace sentences.
Mathematical symbols are used to represent complex formulae or longwinded phrases
which appear repeatedly in the text. By introducing a new symbol, the author avoids
constantly rewriting these formulae (and the reader avoids constantly rereading them). When
used judiciously in this fashion, symbolism can enhance clarity. Mathematical symbols
should only be used in this way.
Introducing new symbols can simplify the exposition, but at a cost: it increases the
burden on the reader’s memory. Only introduce new symbols if the benefit outweighs the
cost.
Any sequence of symbols should read like a fragment of an English sentence (eg. ‘a ∈ B ⊂
C’ means ‘a is an element of B, which is a subset of C’). Thus, mathematical symbols should
be surrounded by enough English words to form a complete sentence. If a mathematical
assertion is nothing but symbols, it is entirely unreadable.
Most important are the stock phrases which indicate the logical role which an assertion
plays in your argument. For example:
Hypotheticals: ‘Assume that...’, ‘Suppose that...’, ‘Let...’, etc.

Consequences: ‘It follows that...’, ‘Thus...’, ‘Therefore....’, ‘But then....’, ‘...so...’, etc.

Logical Connectives: ‘...and...’, ‘...but...’, ‘either....or...’, ‘if....then....otherwise....’, ‘Since.....it

follows....’, etc.

Contradistinctions: ‘On the other hand...’, ‘However...’, etc.

Reminders: ‘Recall...’, ‘Observe that...’, ‘Note...’, ‘Now...’, ‘Remark’, etc.

Desiderata: ‘We want to show that...’, ‘I claim that...’, etc.

Conclusions: ‘We conclude that...’,
Appropriate use of these words can greatly clarify an argument. Omitting them can utterly
obscure it. A common mistake of novices is to write a ‘proof’ consisting of a sequence of
naked mathematical assertions, with not a single word explaining their logical role. This
is impossible to read: it is impossible to know which assertions are hypotheses, which are
consequences of these hypotheses, and which are conclusions.

Equalities
Many novices have a bad habit of ‘proving’ an equality by writing the equality first, and
then ‘working backwards’. While this is sometimes an effective problem-solving strategy,
it is not a correct proof. It gives the reader the impression that you are assuming what
you are trying to prove. For example, suppose we want to prove the trigonometric identity
sec2 (θ) = 1 + tan2 (θ)....
Good:
Bad:
1
sec2 (θ) =
cos2 (θ)
sec2 (θ) = 1 + tan2 (θ)
cos2 (θ) + sin2 (θ)
1 sin2 (θ) =
= 1+ cos2 (θ)
cos2 (θ) cos2 (θ)
cos2 (θ) sin2 (θ)
1 cos2 (θ) sin2 (θ) = +
= + cos2 (θ) cos2 (θ)
cos2 (θ) cos2 (θ) cos2 (θ)
sin2 (θ)
1 cos2 (θ) + sin2 (θ) = 1+
= cos2 (θ)
cos2 (θ) cos2 (θ)
= 1 + tan2 (θ)
1 = cos2 (θ) + sin2 (θ)
1 = 1

Modules
A mathematical proof should be broken down into distinct modules, each of which solves
a particular problem or accomplishes a particular goal. These modules are comparable to
subroutines within a computer program.
Each module should begin by clearly stating its goal. Avoid the ‘abracadabra’ writing
style, where you first present a confusing mass of technicalities, and then at the end, explain
what it was all about. For example:
Bad: Good:
bleah bleah bleah bleah bleah bleah bleah bleah I claim that that all frobnitzes have the type II
bleah bleah bleah bleah bleah bleah bleah bleah Siegel property. To see this, observe that bleah
bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah... Thus we bleah bleah bleah bleah bleah bleah bleah bleah
see that all frobnitzes have the type II Siegel bleah bleah bleah bleah bleah bleah bleah bleah
property. bleah bleah bleah bleah bleah.
A module should have a logical development like a proper English essay: distinct paragraphs,
each developing a distinct idea, and each logically flowing into the next.
Often, a proof has a hierarchical structure, with submodules nested within modules. This
hierarchical structure should be explicitly visible in the page layout. For example:
 Bad:
I claim that that all frobnitzes have the type II Siegel property. To see this, first we must show
that frobnitzes are spurling. Bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah bleah. Next I claim that spurling implies the bi-infinite receding
foobaz condition. bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah bleah. Finally, note that the foobaz condition implies the Siegel
property: bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah.

Good:
Claim 1: All frobnitzes have the type II Siegel property.
Proof:
Claim 1.1: Frobnitzes are spurling.

Proof: Bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah bleah bleah. .......................... 2 [Claim 1.1]

Claim 1.2: Spurling implies the bi-infinite receding foobaz condition.

Proof: bleah bleah bleah bleah bleah bleah bleah bleah bleah bleahbleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah. .............. 2 [Claim 1.2]

Claim 1.3: The Foobaz condition implies the Siegel property

Proof: bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah bleah
bleah bleah bleah bleah bleah bleah. ..................................... 2 [Claim 1.3]

From Claims 1.1, 1.2, and 1.3, we conclude that frobnitzes have the Siegel property 2 [Claim 1]

Intuition pumps
At the beginning, sketch out the strategy of your proof in broad, intuitive terms. This sketch
need not be rigorous, precise, or even mathematically correct (though you must clearly
acknowledge where you bend the truth). The sketch should involve as little notation as
possible. It should provide the reader with a rough ‘mental framework’, upon which to
‘attach’ the technicalities of of the proof. Generally, there are only two reasons why you
would be unwilling to provide such a sketch:

1. You don’t want to reveal your ‘secret insight’, which enabled you to solve the problem.

2. You don’t really understand what you’ve done... you just cobbled together a bunch of
machinery, and it all works, somehow.

Neither of these will endear you to the reader.

Notation Good notation is crucial for effective communication, and depends upon three
principles: Simplicity, Mimesis, and Consistency.

I. Notational Simplicity: Notation should be as simple as possible, while conveying all

essential information. Avoid attaching extra subscripts, superscripts, tildes, hats, primes,
arguments, or other dongles, unless they are strictly necessary to avoid ambiguity.
Sometimes, you must choose between:

(A) Notation which is technically correct, but horribly complex and confusing.
(B) Notation which is technically incorrect, but much simpler to read, and whose meaning
is obvious to anyone with half a brain.

In such a situation, you should always choose (B), but you must explicitly point out that
you are using a ‘technically incorrect’ notation. This is sometimes called ‘abusing notation’.
Some common examples:

• If X and Y are two sets (groups, topological spaces, etc.) and f : X−→Y is an injection
that embeds X as a subset (subgroup, subspace, etc.) of Y, then we sometimes identify
X with its image f (X), and identify each point x ∈ X with its image f (x) ∈ f (X).

• In analysis, one often sees arguments involving sequences, subsequences, subsubse-

quences, etc. The standard ‘reindexing’ trick allows one to avoid introducing subsub-
scripts, subsubsubscripts, etc. For example:
Good:
Bad:
Let {xn }∞n=1 be a sequence in X. Now, X has
Let {xn }∞
n=1 be a sequence in X. Since X has the
the type II Siegel property; so by dropping to a
type II Siegel property, we can find a subsequence
subsequence and reindexing, we can assume that
{xni }∞
i=1 satisfying the bi-infinite receding foobaz
{xn }∞
n=1 satisfies the bi-infinite receding foobaz
condition. Next, by Gromwald’s Other Theorem,
condition. Next, by Gromwald’s Other Theorem,
we can find a subsubsequence {xnij }∞ j=1 which is
we can drop to a subsequence which is strongly
strongly frobnitz. But then Lemma 18(c) yields
frobnitz. By Lemma 18(c), we can drop to a fur-
a subsubsubsequence {xnij }∞ k=1 converging to a
k ther subsequence and reindex, and assume that
fixed point.
{xn }∞
n=1 converges to a fixed point.

II. Notational Mimesis: Notational relationships should reflect mathematical relation-

ships. This mnemonic device helps the reader remember the meanings of myriad symbols.
In particular:

1. Use a specific font to denote objects of a particular type. For example, in linear alge-
bra, you might denote vectors by bold-faced lower-case letters (u, v, w, . . .), matrices
by bold-faced upper-case letters (A, B, C, . . .), scalars by lower-case roman letters
(r, s, t, . . .), and vector (sub)spaces by upper-case ‘blackboard’ font (U, V, W, . . .).

2. Use letters which stand for descriptive words. Hence, ‘f ’ stands for ‘function’; ‘n’
means ‘number’; ‘p’ means ‘prime’; ‘S’ stands for ‘set’; ‘G’ stands for ‘group’, etc.

3. Use alphabetically consecutive letters to denote similar objects. For example, if one
function is called ‘f ’, then the next two functions could be ‘g’ and ‘h’.
Problem: One alphabetical sequence may run into another. For example, if functions
are f, g, h, . . ., and indexes are i, j, k, . . ., then one can’t represent more than three
different functions.

4. Use letters from the same lexicographical family to denote objects which ‘belong’ to-
gether. For example:

(a) If S and T are sets, then elements of S should be s1 , s2 , s3 , . . ., while elements of

T are t1 , t2 , t3 , . . ..
 (b) Use indexing variables which agree with their bounds. Reserve upper-case letters
(eg. J, K, L, M, N, . . .) for the bounds of intervals or indexing sets, and then use
the corresponding lower-case letters (eg. j, k, l, m, n, . . .) as indexes. For example:

Bad: Good:
For all r ∈ [1..n] and q ∈ [1..m], let For all n ∈ [1..N ] and m ∈ [1..M ], let
Xk X̀ XJ X K
A(r, q) = aij (r, q). A(n, m) = ajk (n, m).
i=1 j=1 j=1 k=1

(c) If v is a vector, then its entries should be called v1 , . . . , vN . If A is a matrix,

then its entries should be a11 , . . . , aN M . If V is a vector space, then elements of
V should be v1 , v2 , v3 .
Problem: The entries of vector v1 would be denoted with cumbersome double-
subscripts v11 , v12 , . . . , v1N . There are three solutions:
i. Distinguish elements of V with ‘decorations’, eg. v, v b, v0 , v, v∗ , v† , etc.
e, v
Thus, entries of v e would be denoted ve1 , ve2 , . . . , veN , etc.
Problem: This is impractical when there are more than six or seven distinct
elements. Also, some decorations have special meanings in some contexts.
For example, v0 might mean ‘derivative’, v might be ‘complex-conjugate’; v∗
might be ‘adjoint’; v b might be ‘Fourier transform’; etc.
ii. Distinguish elements of V with superscripts, eg. v(1) , v(2) , . . .. Thus, entries
(1) (1) (1)
of v(1) would be v1 , v2 , . . . , vN . This allows you to represent any number
of elements (including infinite sequences),
Problem: The ‘double indexing’ can be very confusing. Notation with
multiple indices is hard to read, and should be avoided when possible.
iii. Use alphabetically consecutive letters, eg. u, v, w, . . .. Thus, entries of u would
be u1 , . . . , uN . We’ve already mentioned the shortcomings of this method.

Letters from the same lexicographical family should not be used to excess. For example,
a proof involving vectors v, v(1) , v(2) , v(3) , v e(2) , v0 , v
e(1) , v
e, v e0 , v
b, and v can become very
confusing; it would be better to use distinct letters u, v, w, . . ..

III. Notational Consistency: The same notational conventions should be used through-
out the text. Decide at the outset what conventions you will use, and stick to them. For
example, if, on page 1, you use ‘f ’, ‘g’, and ‘h’ for functions, and ‘U’, ‘V’, and ‘W’ for open
sets, then on page 10, do not suddenly switch to ‘φ’, ‘ψ’, and ‘ξ’ for your functions and ‘X ’,
‘Y’, and ‘Z’ for sets.
Whenever possible, conform to the notational conventions established in prior literature.
If everyone else uses ‘φ’ for to mean a morphism and ‘K’ to mean its kernel, do not insist on
using ‘f ’ and ‘X’ instead, unless you have a very good reason. On the other hand, do not
slavishly adhere to stupid conventions that could clearly be improved.

Pictures
A common misconception is that pictures are ‘unmathematical’ or ‘nonrigorous’, and should
be replaced by symbolism whenever possible. In fact, the opposite is true. Most mathe-
matics is motivated by visual intuitions, and most mathematicians think visually. A page
of symbolism is an extremely inadequate substitute for a few good pictures. Math books
(A) (B) (C) (D) φ
X Aa G g A1 A2
B a1 2 eG
A B ψ1 χ ψ2
C

gK
x

K
f
b1 b2 φ ∼
φ
C H B1 B2
eH h

Figure 1:

should be filled with illustrations —at least one per page. They aren’t because publishers
are cheap.
Pictures can never replace a clear written exposition; a proof still needs words. But
pictures can always enhance the clarity of the exposition. Some areas of mathematics (eg.
calculus, linear algebra, geometry, topology) are explicitly geometric in nature, and the value
of pictures is obvious. But pictures are also useful in ‘abstract’ mathematics...

• ...to depict relationships between sets. For example, Figure 1A shows how x ∈ A ∩ B,
C ⊂ B \ A, and A ∪ B ⊂ X. This is called a Venn diagram.

• ...to depict functions. For example, Figure 1B shows that f is a function from A into
B, with image C. Also, f (a1 ) = b1 , and f (a2 ) = b2 .

• ...to depict algebraic structure, by treating groups, rings, etc. as metaphorical ‘vector
spaces’. For example, Figure 1C shows that φ : G−→H is a homomorphism with
kernel K. Here, φ(g) = h, and the preimage of h is the coset gK.

• ...to depict networks of functions between spaces. For example, Figure 1D depicts
functions φ : A1 −→A2 , ψ1 : A1 −→B1 , φe : B1 −→B2 , ψ2 : A2 −→B2 , and χ :
A1 −→B2 , such that ψ2 ◦ φ = χ = φe ◦ ψ1 . This is called a commuting diagram.