LaTeX Authoring Skill

Expert-level LaTeX document authoring following arXiv submission best practices and Overleaf conventions. Produces accessible, semantic, publication-ready documents.

Core Principle

"Use macros according to their meaning, not because of the way they appear visually." — arXiv Best Practices

Semantic markup over visual formatting. Always.

Document Structure

Standard Article

\documentclass[12pt]{article}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{amsmath,amssymb,amsthm}
\usepackage{graphicx}
\usepackage{hyperref}
\usepackage[numbers]{natbib}
\usepackage{booktabs}          % Professional tables
\usepackage{microtype}         % Typography improvements
\usepackage{cleveref}          % Smart cross-references

\title{Article Title}
\author{Author One\thanks{affiliation} \and Author Two\thanks{affiliation}}
\date{\today}

\begin{document}
\maketitle
\begin{abstract}
  Concise summary of the work.
\end{abstract}

\section{Introduction}
\section{Related Work}
\section{Method}
\section{Results}
\section{Discussion}
\section{Conclusion}

\bibliographystyle{plainnat}
\bibliography{references}
\end{document}

Front Matter (arXiv requirement)

Always use semantic front matter commands:

\title{...}
\author{Author One \AND Author Two}
\begin{abstract} ... \end{abstract}

Never simulate titles with

\textbf{\Large Title}

Semantic Markup Rules

DO (semantic)

\emph{key phrase}              % Emphasis with meaning
\textbf{important term}       % Bold with purpose
\section{Introduction}         % Structural heading
\begin{theorem} ... \end{theorem}  % Named environment
\label{eq:energy}             % Reference target
\cref{eq:energy}              % Smart cross-reference

DON'T (visual-only)

{\it key phrase}               % Visual italic, no semantics
{\bf important}                % Visual bold, no semantics
{\Large\bfseries Introduction} % Fake heading, breaks structure
\vspace{1em}                   % Manual spacing hack
\\[1em]                        % Line break as spacing

Mathematics

Inline vs Display

% Inline math
The energy $E = mc^2$ is fundamental.

% Display (numbered)
\begin{equation}
  E = mc^2 \label{eq:energy}
\end{equation}

% Display (unnumbered)
\begin{equation*}
  F = ma
\end{equation*}

% Multi-line aligned
\begin{align}
  \nabla \cdot \mathbf{E} &= \frac{\rho}{\epsilon_0} \label{eq:gauss} \\
  \nabla \times \mathbf{B} &= \mu_0 \mathbf{J} + \mu_0 \epsilon_0 \frac{\partial \mathbf{E}}{\partial t}
\end{align}

Common Math Patterns

% Matrices
\begin{pmatrix} a & b \\ c & d \end{pmatrix}

% Cases
f(x) = \begin{cases} 1 & \text{if } x > 0 \\ 0 & \text{otherwise} \end{cases}

% Operators
\DeclareMathOperator{\argmax}{arg\,max}
\DeclareMathOperator{\softmax}{softmax}

% Fractions (prefer \frac for display, \nicefrac for inline)
\frac{\partial L}{\partial \theta}

% Bold math vectors
\mathbf{x}, \boldsymbol{\theta}

Tables

Professional Tables (booktabs)

\begin{table}[htbp]
  \centering
  \caption{Model comparison on benchmark tasks.}
  \label{tab:results}
  \begin{tabular}{lrrr}
    \toprule
    Model & Accuracy & F1 & Latency (ms) \\
    \midrule
    Baseline    & 0.842 & 0.831 & 12.3 \\
    Our Method  & \textbf{0.917} & \textbf{0.904} & 14.1 \\
    \bottomrule
  \end{tabular}
\end{table}

Never use

\hline

\toprule

\midrule

\bottomrule

Figures

With Alt Text (arXiv accessibility requirement)

\begin{figure}[htbp]
  \centering
  \includegraphics[width=0.8\textwidth, alt={Bar chart showing accuracy improvements from 84% baseline to 92% with our method across three benchmarks}]{figures/results.pdf}
  \caption{Accuracy comparison across benchmark tasks.}
  \label{fig:results}
\end{figure}

Subfigures

\usepackage{subcaption}
\begin{figure}[htbp]
  \centering
  \begin{subfigure}[b]{0.48\textwidth}
    \includegraphics[width=\textwidth]{fig-a.pdf}
    \caption{Training loss}
    \label{fig:loss}
  \end{subfigure}
  \hfill
  \begin{subfigure}[b]{0.48\textwidth}
    \includegraphics[width=\textwidth]{fig-b.pdf}
    \caption{Validation accuracy}
    \label{fig:acc}
  \end{subfigure}
  \caption{Training dynamics over 100 epochs.}
\end{figure}

Bibliography

BibTeX Entry Types

@article{vaswani2017attention,
  title     = {Attention Is All You Need},
  author    = {Vaswani, Ashish and others},
  journal   = {NeurIPS},
  year      = {2017}
}

@inproceedings{devlin2019bert,
  title     = {{BERT}: Pre-training of Deep Bidirectional Transformers},
  author    = {Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
  booktitle = {NAACL-HLT},
  year      = {2019}
}

@misc{anthropic2026claude,
  title        = {Claude 4.6 Model Card},
  author       = {Anthropic},
  year         = {2026},
  howpublished = {\url{https://www.anthropic.com}},
  note         = {Accessed: 2026-04-08}
}

Citation Commands

% natbib
\citet{vaswani2017attention}  → Vaswani et al. (2017)
\citep{vaswani2017attention}  → (Vaswani et al., 2017)
\citep[see][]{vaswani2017attention} → (see Vaswani et al., 2017)

% biblatex
\textcite{vaswani2017attention}
\parencite{vaswani2017attention}

arXiv Submission Checklist

1. Single directory — all files in one flat directory (no subdirectories for figures)
2. Main file — must be compilable with

   pdflatex main.tex

   (or xelatex/lualatex)
3. References — include

   .bbl

   file (pre-compiled bibliography), not just

   .bib

4. Figures — PDF/PNG/JPG only. EPS accepted but PDF preferred. Include alt text.
5. No absolute paths — use

   \includegraphics{figure1}

   not

   \includegraphics{/home/user/fig/figure1}

6. Package compatibility — check LaTeXML supported packages for HTML conversion
7. Compiler — use most recent TeX Live. Set Overleaf to "stop on errors" mode.
8. Font encoding — use

   \usepackage[T1]{fontenc}

   for proper glyph support
9. No custom styles — avoid

   \def

   and

   \newcommand

   that redefine standard commands
10. Comments — remove

    \usepackage{comment}

    sections before submission

Common Mistakes

{\it text}

\emph{text}

\textit{text}

{\bf text}

\textbf{text}

$$display math$$

\[ display math \]

equation

\hline

\toprule

\midrule

\bottomrule

\\[1em]

\vspace

\begin{center}

\centering

\label

\caption

\caption{...}\label{...}

\cite

\citet

\citep

\ref

\cref

\autoref

Document Classes

article

report

book

beamer

standalone

memoir

IEEEtran

acmart

revtex4-2

aastex63

mnras

Essential Packages

amsmath

amssymb

amsthm

graphicx

hyperref

cleveref

\cref

booktabs

natbib

microtype

xcolor

tikz

algorithm2e

listings

minted

-shell-escape

siunitx

\SI{299792458}{\meter\per\second}

subcaption

Package Load Order

Load in this order to avoid conflicts:

\usepackage[utf8]{inputenc}    % 1. Input encoding
\usepackage[T1]{fontenc}       % 2. Font encoding
\usepackage{amsmath,amssymb}   % 3. Math
\usepackage{graphicx}          % 4. Graphics
\usepackage{xcolor}            % 5. Colors (before tikz)
\usepackage{tikz}              % 6. Diagrams
\usepackage{booktabs}          % 7. Tables
\usepackage{algorithm2e}       % 8. Algorithms
\usepackage{listings}          % 9. Code
\usepackage[numbers]{natbib}   % 10. Citations
\usepackage{hyperref}          % 11. Hyperlinks (LAST or near-last)
\usepackage{cleveref}          % 12. Smart refs (AFTER hyperref)

ORCID Integration

ORCID (Open Researcher and Contributor ID) is the standard persistent identifier for researchers. Format:

0000-0002-1825-0097

Including ORCID in LaTeX Documents

\usepackage{orcidlink}  % Provides \orcidlink command

% In author block
\author{
  Jane Researcher\orcidlink{0000-0002-1825-0097} \and
  John Scholar\orcidlink{0000-0001-5109-3700}
}

Manual ORCID (without orcidlink package)

\usepackage{hyperref}
\newcommand{\orcid}[1]{\href{https://orcid.org/#1}{\includegraphics[height=1em]{orcid.png} #1}}

% Usage
\author{Jane Researcher\thanks{\orcid{0000-0002-1825-0097}}}

ORCID in BibTeX

@article{researcher2026,
  author  = {Researcher, Jane},
  title   = {Important Discovery},
  journal = {Nature},
  year    = {2026},
  note    = {ORCID: 0000-0002-1825-0097}
}

ORCID Record Data Types

ORCID tracks: works (papers, books, preprints), employment, education, funding, peer review, research resources, invited positions, distinctions, memberships, qualifications.

ORCID API Integration

• Public API: Read-only access to public researcher data. No auth required.
• Member API: Read/write with OAuth 2.0 permission. Post works, affiliations, funding.
• Sandbox:

  sandbox.orcid.org

  for testing (use before production).
• iD Format: Always display as full URL:

  https://orcid.org/0000-0002-1825-0097

• Schema: v3.0 current. Supports DOI, arXiv ID, PubMed ID, ISBN cross-linking.

Best Practices

• Always collect ORCID iD via authenticated OAuth (never manual entry — prevents typos)
• Display the green ORCID icon alongside the iD per brand guidelines
• Link ORCID to arXiv submissions (arXiv supports ORCID in author metadata)
• Include ORCID in all journal submissions, grant applications, and institutional profiles
• Use the ORCID sandbox (

  sandbox.orcid.org

  ) for development/testing

When This Skill Activates

• Writing LaTeX documents or templates
• Preparing arXiv submissions
• Formatting mathematical expressions
• Creating publication-ready tables and figures
• Managing bibliographies
• Troubleshooting LaTeX compilation errors
• Converting documents between formats
• Creating Beamer presentations
• Integrating ORCID identifiers into papers
• Academic publishing workflow questions