Chapter 3: Document Structure, Syntax and Text

3.1 The smallest complete document

\documentclass{article}          % 1. what kind of document

\usepackage{amsmath}              % 2. preamble: load packages, define things
\title{My First Paper}
\author{Ada Lovelace}
\date{\today}

\begin{document}                  % 3. body: the actual content
\maketitle

\section{Introduction}
Hello, \LaTeX{}! This is a paragraph.

This is another paragraph, because there was a blank line.
\end{document}

Everything before \begin{document} is the preamble: configuration only, no visible output. Everything between \begin{document} and \end{document} is content. Anything after \end{document} is ignored.

3.2 The syntax in five rules

1. Commands start with a backslash: \section, \textbf, \alpha.
   • Mandatory arguments go in braces: \textbf{bold}
   • Optional arguments go in square brackets, before the braces: \section[Short]{A Very Long Title}, \documentclass[11pt,twocolumn]{article}
   • Starred versions: \section*{Acknowledgments} = unnumbered.
   • A command name made of letters eats the spaces after it: \LaTeX is → “LaTeXis”. Use \LaTeX{} is or \LaTeX\ is.
2. Environments wrap a region: \begin{name} ... \end{name}.
3. Groups: braces { } limit scope: {\small this is small} this is not.
4. Comments: % to the end of the line. There’s no block comment by default. Use \iffalse ... \fi or the comment package.
5. Special characters you must escape:

The one that bites ML people constantly: underscores in model names and file names. Write \texttt{train\_model.py}, or use \verb|train_model.py| or \path{}/\url{}.

3.3 Whitespace and paragraphs

• Multiple spaces = one space. A single newline = a space.
• A blank line = a new paragraph. (\par does the same thing.)
• \\ forces a line break without a new paragraph. Avoid it in body text. It’s for tables, addresses, poetry.
• ~ is a non-breaking space. Use it before references so the number never ends up alone at the start of a line: Figure~\ref{fig:a}, Section~3, Dr.~Smith.
• After an abbreviation ending in a period, LaTeX assumes the sentence ended and adds extra space. Fix: e.g.\ this, et al.\ show, or i.e., (a following comma avoids the issue).
• Explicit spacing: \, thin, \quad and \qquad wide, \hspace{1cm}, \vspace{2mm}. Avoid using these as a habit; let the class do layout.
• \newpage, \clearpage (also flushes pending floats).

3.4 Document classes

\documentclass[options]{class}

Common options: 10pt|11pt|12pt, a4paper|letterpaper, twocolumn, draft (shows overfull boxes as black bars, skips images: fast), final.

Conference templates for ML (NeurIPS, ICML, ICLR) use article and add their own style package (\usepackage{neurips_2026}). See chapter 8.

3.5 Packages

\usepackage[options]{package}

A solid default preamble for a research paper:

\usepackage[utf8]{inputenc}      % harmless; UTF-8 is the default since 2018
\usepackage[T1]{fontenc}         % proper font encoding (hyphenation of accented words, copy-paste)
\usepackage{lmodern}             % or: times, mathptmx, newtxtext/newtxmath, libertine, ...
\usepackage{microtype}           % subtle spacing improvements: always use it
\usepackage{amsmath,amssymb,amsthm,mathtools}  % math (chapter 4)
\usepackage{graphicx}            % \includegraphics
\usepackage{booktabs}            % good tables
\usepackage{xcolor}              % colors
\usepackage{hyperref}            % clickable links (load late)
\usepackage[capitalize,noabbrev]{cleveref} % smart references (load after hyperref)

Package load order matters a bit. Rule of thumb: hyperref goes near the end, and cleveref comes after hyperref. Conference style files often load some of these already. Loading a package twice with the same options is harmless; with different options it gives an “option clash” error.

Find package docs: texdoc booktabs in the terminal, or ctan.org/pkg/booktabs.

3.6 Title, authors, abstract

\title{Attention Is All You Need (Again)}
\author{
  Ada Lovelace\thanks{Equal contribution.} \\
  Analytical Engine Lab\\
  \texttt{ada@example.org}
  \And                          % \And / \AND are NeurIPS-style separators; plain article uses \and
  Alan Turing\footnotemark[1] \\
  Bletchley Institute
}

\begin{document}
\maketitle

\begin{abstract}
We propose ...
\end{abstract}

Author formatting is the most template-specific thing in LaTeX. ICML uses \icmlauthor{}, IEEE uses \IEEEauthorblockN, acmart uses \author{}\affiliation{} blocks. Always copy the pattern from the template’s example file.

3.7 Sectioning

\part{...}            % rarely used in papers
\chapter{...}         % report/book only
\section{Introduction}
\subsection{Motivation}
\subsubsection{A detail}
\paragraph{Bold run-in heading.} Text continues on the same line.
\subparagraph{...}

\section*{Acknowledgments}   % unnumbered, not in table of contents

\appendix                    % from here on, sections are lettered A, B, C...
\section{Proofs}

\paragraph{} is heavily used in ML papers for the bold inline mini-headings (“Datasets. We evaluate on…”).

Table of contents: \tableofcontents (rare in papers, common in theses and long appendices).

3.8 Text formatting

Old-style {\bf ...} and {\it ...} are deprecated, but you’ll still see them in old papers.

Font sizes (as switches): \tiny \scriptsize \footnotesize \small \normalsize \large \Large \LARGE \huge \Huge. Example: {\small text}.

Quotes: use ``double'' and `single'. Never type " for quotes in pdfLaTeX: you get a straight or closing-style quote on both sides instead of proper opening quotes. Or use the csquotes package with \enquote{...}.

Dashes:

• - hyphen: state-of-the-art
• -- en dash: ranges pages 10–20, 2019–2024, encoder–decoder
• --- em dash: a break in thought—like this

Ellipsis: \ldots (text) or \dots. Not three periods.

Colors: \textcolor{red}{text}, {\color{blue} text} (package xcolor). Define your own: \definecolor{mygreen}{RGB}{0,150,80}.

3.9 Lists

\begin{itemize}          % bullets  (Markdown: - item)
  \item First
  \item Second
  \begin{itemize}
    \item Nested
  \end{itemize}
\end{itemize}

\begin{enumerate}        % numbered (Markdown: 1. item)
  \item First
  \item[(a)] Custom label
\end{enumerate}

\begin{description}      % term–definition
  \item[Encoder] maps inputs to latents.
  \item[Decoder] maps latents to outputs.
\end{description}

The enumitem package lets you control spacing and labels. Use it to save space in papers:

\usepackage{enumitem}
\begin{itemize}[leftmargin=*, nosep]
  \item Compact list, no extra vertical space.
\end{itemize}
\begin{enumerate}[label=(\roman*)]   % (i), (ii), (iii)
  \item ...
\end{enumerate}

The classic “our contributions are:” list in ML papers is an itemize with nosep or itemsep=0pt.

3.10 Footnotes, quotes, verbatim

Transformers\footnote{Introduced by \citet{vaswani2017}.} are everywhere.

\begin{quote}
  A short block quote.
\end{quote}

\begin{verbatim}
Text exactly as typed, including _ % $ & characters.
\end{verbatim}

3.11 Labels and cross-references

This is one of LaTeX’s superpowers.

\section{Method}\label{sec:method}

\begin{equation}
  y = Wx + b \label{eq:linear}
\end{equation}

\begin{figure}
  ...
  \caption{Our architecture.}\label{fig:arch}   % \label AFTER \caption!
\end{figure}

As shown in Section~\ref{sec:method}, Eq.~\eqref{eq:linear}, and Figure~\ref{fig:arch}...
On page~\pageref{fig:arch}...

Conventions for label prefixes: sec:, subsec:, eq:, fig:, tab:, alg:, thm:, lem:, app:. They’re just text, but they keep you sane.

cleveref writes the word for you:

\cref{fig:arch}          % → "Figure 3"   (with [capitalize,noabbrev] options)
\Cref{eq:linear}         % → "Equation (2)", capitalized for start of sentence
\cref{fig:a,fig:b,fig:c} % → "Figures 1 to 3"

Most modern ML papers use \cref. (Chapter 7 covers hyperref + cleveref setup.)

Gotcha: \label refers to the most recent numbered thing. In figures, it must come after \caption, or you get the section number instead.

3.12 Defining your own commands (macros)

This is where LaTeX beats Markdown: you can define your own vocabulary.

% In the preamble (or macros.tex)
\newcommand{\method}{\textsc{FastNet}}           % no arguments
\newcommand{\R}{\mathbb{R}}
\newcommand{\norm}[1]{\left\lVert #1 \right\rVert}    % one argument: #1
\newcommand{\inner}[2]{\langle #1, #2 \rangle}         % two arguments
\newcommand{\vect}[1][x]{\mathbf{#1}}             % optional arg with default "x"

\renewcommand{\vec}[1]{\boldsymbol{#1}}           % redefine an existing command

% xspace: adds a space after a macro unless followed by punctuation
\usepackage{xspace}
\newcommand{\ours}{\textsc{FastNet}\xspace}

Usage:

\method outperforms baselines.        % careful: eats the space → "FastNetoutperforms"
\method{} outperforms baselines.      % OK
\ours outperforms baselines.          % OK with \xspace
Let $x \in \R^d$ and $\norm{x} \le 1$, with $\inner{u}{v} = 0$.

Changing \method in one place renames your method across the whole paper. That’s invaluable when reviewers or co-authors want a different name.

Collaboration/TODO macros (very common)

\newcommand{\todo}[1]{\textcolor{red}{[TODO: #1]}}
\newcommand{\ada}[1]{\textcolor{blue}{[Ada: #1]}}
\newcommand{\alan}[1]{\textcolor{orange}{[Alan: #1]}}

% to hide all notes for the final version:
% \renewcommand{\todo}[1]{}

Or use the todonotes package for margin notes: \todo{fix this}.

Environments of your own

\newenvironment{keyidea}
  {\begin{quote}\itshape\textbf{Key idea.}~}
  {\end{quote}}

3.13 Boxes, minipages and columns

\begin{minipage}{0.48\textwidth}
  Left content
\end{minipage}\hfill
\begin{minipage}{0.48\textwidth}
  Right content
\end{minipage}

\fbox{framed text}
\parbox{5cm}{a paragraph in a fixed-width box}

Useful length units: \textwidth (text block width), \linewidth (current line width, use this inside columns/minipages), \columnwidth, pt, mm, cm, in, em (width of an “M” in the current font).

For fancy colored boxes (like “takeaway” boxes in papers), use tcolorbox:

\usepackage{tcolorbox}
\begin{tcolorbox}[colback=blue!5, colframe=blue!60!black, title=Takeaway]
  Scaling data matters more than scaling parameters.
\end{tcolorbox}

(blue!5 means 5% blue mixed with white. That xcolor syntax shows up everywhere.)

3.14 Markdown → LaTeX quick map

3.15 Key takeaways

• Preamble = setup; body = content. Commands \x{}, optional args [ ], environments \begin{}...\end{}.
• Escape % $ & # _ { } ~ ^ \. Underscores in code names are the usual trap.
• Blank line = paragraph. Use ~ before refs. Avoid \\ in prose.
• \label + \ref/\cref for everything. Label after caption.
• Define macros for your method name, notation and TODOs.