unicodemath

Parse UnicodeMath and AsciiMath, the two main linear (plain-text) math notations, into a common tree, then emit LaTeX or anything else. UnicodeMath is the notation Microsoft Word uses natively for equations (Unicode Technical Note 28), so ∑_(i=1)^n i^2 or a^2+b^2=√(c^2) is valid input. AsciiMath is the pure-ASCII cousin: sum_(i=1)^n i^2, AA x in RR x^2>=0.

The parser is a small Pratt parser, a clean-room Python port of the design of Peter Jipsen’s umath2latex.js. Its LaTeX output byte-matches that reference implementation across the sample set published with it.

Install

pip install unicodemath

Use

from unicodemath import parse, to_latex

to_latex(parse('∀x>0 x^a x^b = x^(a+b)'))
# '\\forall x > 0(x^ax^b = x^{a+b})'

to_latex(parse('int_0^oo e^(-x) dx', dialect='asciimath'))
# '\\int_0^\\infty e^{-x}dx'

Matrices work in both dialects: UnicodeMath ■(a&b@c&d) (and █(...) for an equation array, which becomes aligned), AsciiMath [[a,b],[c,d]].

to_latex(parse('[[a,b],[c,d]]', dialect='asciimath'))
# '\\begin{bmatrix}a&b\\\\c&d\\end{bmatrix}'

prefixform renders the parse tree in a compact op(arg,...) form, useful for checking how an expression was read:

from unicodemath import prefixform
prefixform(parse('|x+y| ≤ |x|+|y|'))
# '≤(|(+(x,y)),+(|(x),|(y)))'

Backends

parse returns a tree of Node objects, each with sym (the source symbol), kind (how it was constructed: infix, chain, prefixop, around, matrix, …), and children in arg, arg2, arg3. A backend is a function from that tree to a string; to_latex is the one included, and dispatches on kind. New dialects register a tokenizer in unicodemath.core.TOKENIZERS.

Scope

The goal is the common mathematical core of both notations, not either spec in full. Known gaps: AsciiMath text() and accents (hat, vec, …) are not supported, and greedy tokenizing means multi-letter names the symbol table doesn’t know (like sinh) parse as juxtaposed letters. UnicodeMath control words (\alpha style) pass through as literal LaTeX names.

Development

Version lives in unicodemath/__init__.py as __version__; release uses fastship (ship-bump, ship-gh, ship-pypi). Tests: pytest -q. The Jipsen sample fixtures in tests/fixtures/ were generated one time from the reference JS and are the port contract: our output must byte-match them.