How to Write a Math Paper (Rough Draft)
J. C. Lagarias
February 1, 2006
Abstract
This is a skeletal and opinionated description of paper writing. Take
from it the good things, and disregard the bad things.
1 First Draft
The hardest thing about writing a paper (in my experience) is getting a first
draft written.
In particular, if there are proofs, the hard part means getting some draft of
a proof written down on paper, not necessarily with every detail included.
It is not necessary to get a polished presentation on the first draft.
After a first draft is written, one can edit and re-edit the paper to get it in
a better form. Editing tools now permit significant rearrangement/reordering
of contents. (Latex will keep track of renumbering everything, section numbers,
theorem numbers, page numbers.)
Many of the principles of good writing in general apply to math paper writ-
ing. I wish I knew what they all were. However the next section gives the point
more important for math papers than others; good organization.
2 Organization
A paper should tell a story, if possible. This makes the paper easier to follow.
This is true also of “experimental” papers, where the paper may describe the
journey.
An example of a well-organized expository paper, that tells a story, is one
by Andrew Granville and Greg Martin [1].
I think a paper lives or dies in its organization. If the organization is good,
revision of the paper lies locally in the details. (This is a “great truth”. The
1
difference between a “truth” and a “great truth” is that a “great truth” is one
whose opposite is also a “great truth.”)
The paper should have pointers explaining where the paper is going. For
example, it is hard to read a proof that proceeds line by line, without a clue
where the author is going. (Unfortuately a large number of math papers are
written in this fashion.) I am not sure how to explain this further, except by
example.
The idea is that the logical flow of the paper should carry the reader along.
As a question is raised in their mind, the paper is ready to answer it, or ac-
knowledge it as an unanswered question.
However math papers are (usually) not detective novels, Where the paper
is going should be explained in the introduction (at the end). The main results
usually should be stated near the beginning, possibly in a section 2 titled main
results, if they are too complicated or techical for easy statement in the intro-
duction.
One viewpoint is, that if the entire paper were destroyed except for the intro-
duction, it should still be possible to have a good idea what the paper was about.
For example, the Archimedes Palimpsest, (re?)discovered in 1908, contains
versions of “The Method of Mechanical Theorems”, “On Floating Bodies” and
a fragment of “The Stomachion”. What is “The Stomachion” about? No one
is quite sure, as the fragment cuts off at this point.
Remark There is a virtually inexhaustible supply of badly written math pa-
pers. However if a paper has a really tremendous result, it can transcend the
writing style of the author.
3 Introduction
The introduction sets the stage for the paper.
(1) It states the problem, the history of work on this problem and related
problems.
(2) It gives pointers to relevant prior work [4]. and sometimes peripherally
related prior work [2], [3].
(3) It states the main results, either formally as a theorem or informally,
with a pointer to where the main result appears.
2
(4) It states open questions. These can be stated as questions, if one has
no idea of the outcome and wishes to play it safe, or conjectures if one is very
confident of the outcome.
Question. How long would it take 50000 monkeys, typing on typewriters at
the rate of one character per second, for one of them to reproduce in order the
complete works of Shakespeare?
For results one has accumulated significant evidence in favor, one can for-
mulate a “guess.”
Conjecture. (“Ramanujan”) The Riemann zeta function has too many ze-
ros.
[The refers to G.H. Hardy’s remark that much of Ramanujan’s work on prime
numbers would have been completely correct if the Riemann zeta function had
no complex zeros].
[Good conjectures can enhance one’s reputation. Bad conjectures, meaning
ones that are proved false, cast a black mark. Actually I prefer to believe the
shade of the mark varies-as some of my own conjectures have been disproved.]
Remarks. (1) Most of the general bibliographic citations are in the introduc-
tion. In the body of the paper, refences will often be in proofs, to shorten work.
(2) The reliability of the author(s) can sometimes be assessed with how care-
fully, extensively and well chosen the bibliography is. A narrow bibliography,
particularly restricted to that of the author, is suspicious. (Not always. Benoit
Mandlebrot, a great applied mathematician/physicist is fond of citing mostly
his own work.)
4 How Many Sections?
You can subdivide your paper into several sections. This is done with the aim
of clarity of organization. If you are exploring several different questions, you
can have one section for each topic.
You might have a separate section for Computational Issues. This might
describe how to write an efficient program to compute something. (Computer
code might be included in an Appendix.)
There can be a concluding section on further directions of work, open prob-
lems remaining.
3
5 Definitions, Theorems and Proofs.
Definitions are an important part of a paper. They are framing the discourse.
They can be set off, formally, for example:
Definition 1. A point is that which has no part.
[This is the first definition in Euclid’s “Elements”.]
A lot of work in mathematics is in the choice of the “correct” definition.
Good definitions encode (and hide) some of the difficulties of the problem. (In
this regard, the definition of “point” is not very clear. Hilbert argued that if
geometry had the correct foundations, it should be possible to read the proofs
with “Point, Line, Plane” replaced with “Table, Chair, Beer Mug”. Well, you
could do this, but the text would be hard to read.]
The standard proof style is something like this.
Theorem 1. All horses are red.
Proof. This result is proved by induction on the number of horses n, starting
with n = 1.
Induction Step. Suppose it is true for n, to show it is true for n + 1. Take
the n + 1 horses, and form a group of n of them. By the induction hypothesis,
they are all red. Now swap one of these horses this group with the remaining
horse. Again one has a group of n horses, all of which are red by the induction
hypothesis. Since all n + 1 horses are now known to be red, the induction step
is complete. ¤.
Remarks. (1) This proof is an example of the dictum: “Don’t get it right,
get it written.” The base case of the induction has not been checked, and is
false. The induction step is fine.
(2) One can debug theorems, as one debugs code, by careful staring at the
written form. Unfortunately proofs are written in a high-level language (often
English). No compiler is currently available to check if it is a legal proof, be-
cause proofs are generally not written down in a formal language.
6 Anecdotes
(1) A story told about André Weil, and his book “Foundations of Algebraic
Geometry.” This was studied in a reading course at Johns Hopkins University
4
in the early 1950’s, and the students complained it was too dry. The instructor
mentioned this to Weil, on seeing him at the Institute for Advanced Study. His
reply: “I suffered writing it. Why shouldn’t they suffer reading it?”
(2) From Littlewood’s “Miscellany”[5, p. 57]:
“A published paper had near the beginning the passage: ’The object of
this paper is to prove (something very important)’ It transpired, with great
difficulty, and not till near the end, that the ’object’ was an unachieved one.
(The author intended no dishonest claim, but his use of language was unusual.)”
(3) It is always good to make a reader want to read the paper you are writ-
ing. “How many psychiatrists does it take to change a lightbulb?” “Only one,
but the lightbulb has to want to be changed.”
Good luck with your math paper writing.
References
[1] A. Granville and G. Martin, Prime Number Races, Amer. Math. Monthly
113, No. 1, (2006), 1–33.
[2] J. C. Lagarias, The 3x + 1 problem and its generalizations, Amer. Math.
Monthly 92 (1985), 3–21.
[3] J. C. Lagarias, The 3x + 1 Problem: An annotated bibliography, eprint:
arxiv:[Link]/0309224, v5
[4] J. C. Lagarias, How to write a math paper (0-th draft), unpublished.
[5] J. E. Littlewood, Littlewood’s Miscellany, Edited and with a foreword by
Bela Bollobas, Cambridge University Press; Cambridge, UK 1986.