latex2unicode.py

latex2unicode.py addresses a major pain point in scientific web publishing: rendering TeX math expressions without heavyweight JavaScript libraries or font dependencies.

Pathbuild/latex2unicode.py

LanguagePython

Lines368

Sourcebuild/latex2unicode.py
at 406d3e423

Read this when Use this page when tracing Python helper scripts for metadata cleanup, text processing, PDFs, dates, or generated content around latex2unicode.

Overview

latex2unicode.py addresses a major pain point in scientific web publishing: rendering TeX math expressions without heavyweight JavaScript libraries or font dependencies. While full LaTeX rendering (via KaTeX or MathJax) is necessary for complex equations, simple expressions like $X$ or $n^2$ can be converted to native HTML like X or n2, saving megabytes of assets and rendering instantly.

The script uses GPT-5.4-mini to convert inline TeX from ArXiv-style abstracts into Unicode symbols, HTML tags (, , ), and minimal CSS. It handles Greek letters (α, β, Δ), mathematical operators (∫, ∑, ∇), special fonts (𝒪, ℝ, ℂ), and complex structures like superimposed sub/superscripts using a custom .subsup CSS class.

The conversion rules are extensive and nuanced: use FRACTION SLASH (⁄) for small fractions like 1/2 → 1⁄2, but BIG SOLIDUS (⧸) for symbolic fractions a/b → _a_⧸_b_. Skip complex structures like matrices, TikZ diagrams, or deeply nested fractions—those require full LaTeX rendering.

Key Functions

Prompt construction: 250+ line prompt with detailed conversion rules and examples
OpenAI API call: Uses gpt-5.4-mini with system prompt as "skilled mathematician & tasteful typographer"
Output linting: Post-validates the output (balanced tags, no digits/operators/punctuation inside , no  before  in .subsup); on lint failure it retries once, then emits the input unchanged to stdout and exits non-zero
Output handling: Strips trailing newlines to preserve inline text flow

Command Line Usage

# Convert inline TeX from stdin
echo 'O(1)' | OPENAI_API_KEY="sk-XXX" python latex2unicode.py
# Output: 𝒪(1)

# Convert from command line argument
OPENAI_API_KEY="sk-XXX" python latex2unicode.py '\\frac{1}{2}'
# Output: 1⁄2

# Complex example with subscripts/superscripts
echo '\\int_a^b f(x) dx' | OPENAI_API_KEY="sk-XXX" python latex2unicode.py
# Output: ∫<span class="subsup"><sub><em>a</em></sub><sup><em>b</em></sup></span> <em>f</em>(<em>x</em>) <em>dx</em>

# Natural language input also works (bonus feature)
echo 'asymptotically square root n' | OPENAI_API_KEY="sk-XXX" python latex2unicode.py
# Output: 𝒪(√<em>n</em>)

When NOT to convert:

Block equations or complex structures (matrices, TikZ)
Custom operators or redefined commands
Expressions needing special positioning (overrightarrow, underline)

Overview​

Key Functions​

Command Line Usage​

Overview

Key Functions

Command Line Usage