Acronyms and symbol lists

Long documents accumulate abbreviations, and the conventions around them are strict: an acronym is spelled out in full at first use, abbreviated afterward, and a thesis usually needs a list of abbreviations in the front matter. Maintaining that by hand fails in a predictable way. You reorder two chapters, the first use moves, and suddenly “ML” appears three pages before “machine learning (ML)” does. The glossaries-extra package tracks first use for you.

The automated route

\usepackage[acronym]{glossaries-extra}
\makeglossaries
\newacronym{ml}{ML}{machine learning}

The acronym option enables acronym support, and \makeglossaries switches on the machinery that collects entries during the compile. Each \newacronym takes three arguments: a key you will type in the text, the short form, and the long form. Define all of them in the preamble or in a separate file you \input, so they live in one place.

In the body, you never type the acronym itself:

\gls{ml} on first use expands; later uses stay short.
\printglossaries

The first \gls{ml} in the document prints “machine learning (ML)”, and every later one prints just “ML”. Because the package tracks usage rather than position in the source, the expansion follows the first use wherever your restructuring moves it. \printglossaries prints the collected list of abbreviations wherever you place it, typically after the table of contents. Related commands cover the grammar around this: \glspl for plurals, \Gls for a capitalized start of sentence.

The catch is the build. Requires a glossaries build step: between LaTeX passes, the makeglossaries helper must run to sort and format the entries, or \printglossaries silently produces nothing. Full latexmk setups handle this automatically; simpler pipelines may not, and glossaries-extra offers a fallback for them. If your list stubbornly refuses to appear, the missing helper run is the first suspect, well ahead of your document.

The hand-built route

For a simple list of symbols, a hand-maintained description environment often suffices:

\section*{Notation}
\begin{description}
 \item[$\eta$] learning rate
 \item[$\theta$] model parameters
\end{description}

Each \item[...] prints its bracket argument as a bold label with the explanation following, which is exactly the shape of a notation list. There is no first-use tracking, but a symbol list rarely needs any; symbols are defined where they first appear in the math regardless.

The honest advice is to match the tool to the scale. A paper with five acronyms is easier to manage by hand. A 200-page thesis with forty is not, and there the package pays for its setup cost many times over. The lists lesson covers the description environment in more detail.