LaTeX Best Practices

If you are a physicist, chances are good that you are using LaTeX to write texts and typeset your formulas. If you are not, you should (the argument is made elsewhere).

But how do you write good LaTeX? Writing a good paper is one thing, but writing good LaTeX is another story. There are quite a few blog posts on how not to write LaTeX (here for example), but that is not too constructive.

This post will not be a huge explanation on LaTeX internals, but rather a collection of best practices with a motivation. If you have ideas for additions or other opinions, please write them in the comments.

So let’s get started:

  1. Stick to convention throughout the document
    Conventions help to make the LaTeX source code more readable. If you start labeling your figures or equations with a certain convention (see below), stick with the same convention throughout the document. It will make late changes much easier.

  2. One sentence per line in the tex-source
    Many software development tools (or Linux tools) work on a line-by-line basis. LaTeX does not render single return characters in the final PDF, only a free line (two return characters) result in a new paragraph. If you stick with the one-sentence-per-line policy in your LaTeX source, you can use all of them. Some examples are meld, diff or vimdiff. These tools do comparisons on a line-by-line basis and merging a single sentence is much easier than merging a full paragraph.

  3. Use align not eqnarray
    The eqnarray environment in LaTeX is deprecated and its typesetting is problematic (for details look here and here). The equation environment is fine for single-line equations. If you want a single, simple rule: always use align for equations.The eqnarray environment in LaTeX is deprecated and its typesetting is problematic (for details look here and here). The equation environment is fine for single-line equations, but does not work for multiple-line equations. If you want a single, simple rule: always use align for equations. Within the align environment you can use \label{eq:label} for every single line, or use \nonumber at the end of a line to avoid equation numbers, for example when you need to break one long equation into two lines. You can align the two lines in the align environment by placing the &-symbol (you can use multiple &-symbols in one line if it needs to be aligned at multiple places, but note that the align environment has an alternating right/left alignment, meaning that sometimes && may be a better option, see this post for more explanation).

  4. Define commands with reasonable names
    Often you have to use a certain symbol multiple times in a document and you would like it to mean the same thing. Consider for example the double-script R, $\mathbb{R}$ . You can, of course, write \mathbb{R} a lot of times, but wouldn’t it be easier to read if you wrote what you meant? If you define instead the command reals as \newcommand{\reals}{\mathbf{R}}, you can just use the much more readable command \reals in your equations. If you want a math command to work both in text and equations, you can wrap the math command in \ensuremath{}. You don’t have to do this for full symbols, but you can also just define an addition to a symbol to make equations more readable. Like the adjoint operator, frequently used in quantum mechanics, \newcommand{\dgr}{^\dagger}

  5. Use the physics package
    The physics package is an excellent collection of useful LaTeX commands. Starting with matrices and vectors, symbols for partial differential equations, all the way to quantum mechanical notation. Here is just a little taster:

  • \ket, \bra are more semantic than \langle,\rangle. Do not use < and > for expectation values!
  • \mqty for matrices and vectors, using the usual LaTeX table notation

If you want to know more, have a look at the package documentation.

  1. Use \eqref, not (\ref{some_label})
    While \ref is creating references to tables and figures, use \eqref to reference equations. It will directly add the correct parentheses to the number and it is more readable for your collaborators. They know that you are referencing in equation when they are reading the source.

  2. Prefix your labels
    Scientific texts contain many references to figures, sections, tables and equations. By prefixing the label of an object with eq (equations), sec (sections), fig (figures) or tab (tables). A section label for the introduction might look like \label{sec:introduction}. While writing, you will be clear on whether you are referencing the correct type of object. If you want to make references even easier, consider having a look at the cleveref package or a post about it here.

  3. Curiosities about symbols
    Some composed symbols are actually available as a single symbol to make sure that the typesetting is correct. Here are a few examples

  • Use \coloneqq , not “:=”
  • Use \colon in formulas, not “:”
  • Use \dots , not “…’’ (three dots)
  • Use \sin() (\arccos(), etc.), not “sin()” (“arccos()”, etc.), the brackets will scale automatically with the argument. For a list of all symbols and operators available, see this wiki-page
  1. Write your units in text-format, not in equation-format
    If you are writing equations involving masses $m$ and lengths of unit m (meter), using the proper formatting is helpful for distinguishing the two. Generally, the variables are written in equation-style, while units are in text-style. This can get tricky when writing, for example, kg s /m^2. One use \text{} to write the units within an equation or, use $^2$ after the in-text kg s/m. However, if this gets too tedious there is also a package that allows for a more clear writing style, the siunitx package. It helps you to write units directly in an equation-environment like: \si{\kilogram\second\per\metre\squared}. You can also use shorter notation like \s for \second and play with the style in which it appears with \si[per-mode=symbol]{}, see this page for the full manual.

I would love to treat this document as a growing compilation of LaTeX best practices. If you have other best practices or have a different opinion, feel free to comment below.

This blog post was written together with Martine Schut, PhD student in Groningen. Thanks to David Dechant for spotting several typos.

Patrick Emonts
Patrick Emonts
Postdoctoral Researcher

My research interests include tensor networks, lattice gauge theories and quantum information.