LaTeX for Creating and Annotating Figures, by Jochen Autschbach

[Home] [GRP] [RES] [PUB] [ETC] [LEC] [S/W]

Using LaTeX to Create and Annotate Figures

In my (Jochen’s) ﬁeld of science, it is often necessary to combine many diﬀerent kinds of graphics into single ﬁgures with sub-panels and added labels, arrows, or other items. These kinds of ﬁgures are then used in research publications or proposals, for conference presentations or posters, or on a web page. An example for such a ﬁgure is this

Furthermore, to avoid making silly looking ﬁgures (e.g., ﬁgures with much too large labels) for journal publications or books, it is usually a good idea to create graphical material in a speciﬁc target size, with matching font sizes being not too small and not too large. [Credit to Szabolcs Horvát, the author of the excellent ‘MaTeX’ package for Mathematica, who made a good case for designing ﬁgures in the correct size (see MaTeX documentation).] Sometimes, we need to create labels in such ﬁgures that would normally require some kind of equation editor. After weighing the pros and cons of various ways to create such ﬁgures, I have settled (after more than 2 decades of trying diﬀerent things) on an approach using LaTeX. To be clear, the procedure outlined here is about making stand-alone PDF ﬁles of appropriately sized and labeled compound ﬁgures. These PDFs can be included easily in a LaTeX manuscript but also in a word-processing software of your choice. (See the end of this page for format conversion tips.)

For example, I recently needed to make a ﬁgure combining a molecular orbital (MO) diagram with several MO visualizations (as isosurface plots) and assorted labels, like the one shown above. The MO visuals were generated with a separate software in comparatively high resolution (on the order of 1500x1500 pixels in the example). Speciﬁcally, these images were generated in PNG format with one of my Mathematica notebooks, using volume data ﬁles for the diﬀerent MOs that were created based on calculations with a quantum chemistry program package. However, such isosurface plots can be created with a large variety of software packages, open-source or proprietary.

In this archive (tar format, 9MB), in directory plots/, there is a ﬁle panel-mo-diagram-and-orbitals.tex used to assemble the various graphical objects and labels, and a shell script make-panels.sh to compile the ﬁgure with xelatex. There is furthermore a ﬁle 10-modiagram.tex that is ﬁrst compiled by the script to prepare the MO diagram, which is subsequently included in panel-mo-diagram-and-orbitals.tex. Both LaTeX ﬁles can also be compiled individually with xelatex, which is useful during the trial & error period of designing the combined ﬁgure. The script is written such that tex ﬁles with names starting with a double-digit number (01, 10, 23, 99) get compiled ﬁrst, in the order of numbering, and then tex ﬁles whose names start with panel- get compiled subsequently in the order of how your system sorts them in a ﬁle listing. This was done so that dependencies of a given panel are reliably compiled ﬁrst. The script will stop whenever there is an error encountered. LaTeX errors will be listed in the resulting .log ﬁles.

Before you try out this example, please make sure you have a working TeX distribution with xelatex and STIX2 (or STIXTwo) and various other font sets, and Ghostscript. If you can successfully compile my article template posted here (direct download), you should be good to go as far as font installations are concerned (or switch to default fonts). To compile the panel example, your TeX installation also needs a variety of utility packages such as tikz and pgfplots. As a Linux user, I usually install a TeX Live distribution including all packages on my systems. Other TeX distributions have similar ‘full’ installation options. (Who cares about a few GB of extra disk space usage these days?)

In each tex ﬁle in the archive, not far below \begin{document} there is code to set the size of the plotting canvas in centimeters. For example, in panel-mo-diagram-and-orbitals.tex we set the ﬁgure canvas initially to be 16 cm wide and 5.5 cm high. This is a suitable size for a ﬁgure that would span up to two columns in a typical scientiﬁc journal. The code is

\setlength{\mx}{16cm} 
\setlength{\my}{5.5cm}

This can be adjusted, of course. Next, closer to the end of the ﬁle there is some code that will create a grid with 0.5 cm spacing the size of the canvas speciﬁed earlier, which you can use later to aid placement of images, labels, and so on. Using

% Grid to help placing objects, if so requested 
\renewcommand{\showgrid}{true} % use true or false

with a value of true draws the grid. Set to false and recompile, and the grid will not be shown. Note that with the standalone documentclass, the resulting ﬁgure will be cropped automatically, so if the grid extends beyond any of the drawing elements then the ﬁgure without the grid will be smaller than the canvas size speciﬁed.

Without having anything placed in the panel, with \showgrid set to true, the panel (after compiling panel-mo-diagram-and-orbitals.tex with xelatex looks like this:

The LaTeX code for the MO diagram is kind of complicated (though relatively easy to use) and was put together with the help of many posts on Stack Exchange. It has an initial canvas setting of 8.5 x 5 cm but ends up being smaller than that after the grid is removed. Since we are creating the ﬁgure in the intended size, with matching font sizes, the MO diagram does not need to be scaled when it is included in the panel. After it has been compiled into a PDF, we include it in panel-mo-diagram-and-orbitals.tex with the code

% orbital level plot: 
\node[above right] (image) at (0.5,0) { 
  \includegraphics{10-modiagram} 
};

The .pdf ﬁle name extension is not required for xelatex, but make sure the PDF is the only graphics version of the MO diagram, unless you are sure that the PDF version would be included even if a PNG or some other format version is present. I note in passing that JPEG is not a good format for line drawings such as function graphs or MO diagrams and other types scientiﬁc graphics that tend to have a lot of white space and even colors. Anyhow, the resulting compiled panel now looks like this:

(Of course, one can draw the MO diagram with a diﬀerent software and include the resulting image ﬁle in the panel instead.) Next, we place symmetry labels for the MO levels in the diagram. The color specs and the macros \ttwog (typeset: $t_{2 g}$ ) etc. are deﬁned in the tex ﬁles. I forgot why there is a \strut in each of the labels but it has to do with the placement and the vertical extension of the box holding the label. We add the code

\node[right] at (5,4.05){\strut \orange{\eg{}}}; 
\node[right] at (5,3.4){\strut \green{\ttwog{}}}; 
\node[right] at (5,2){\strut \red{\toneu{}}}; 
\node[right] at (5,0.4){\strut \blue{\ttwou{}}}; 
\node[right] at (5,0.1){\strut \purple{\atwou{}}};

The coordinates used in those commands correspond to the location within the grid that we drew on the canvas. The panel now looks like this:

Next, we place a set of nine MO isosurface images, using

% orbital plots 
\node (image) at (7.5,3.5) { 
  \includegraphics[scale=0.1]{U_t2g_ADF} 
}; 
\node (image) at (9,3.5) { 
  \includegraphics[scale=0.1]{Np_t2g_ADF} 
}; 
\node (image) at (10.5,3.5) { 
  \includegraphics[scale=0.1]{Pu_t2g_ADF} 
}; 
 
\node (image) at (7.5,2) { 
  \includegraphics[scale=0.1]{U_t1u_ADF} 
}; 
\node (image) at (9,2) { 
  \includegraphics[scale=0.1]{Np_t1u_ADF} 
}; 
\node (image) at (10.5,2) { 
  \includegraphics[scale=0.1]{Pu_t1u_ADF} 
}; 
 
\node (image) at (7.5,0.5) { 
  \includegraphics[scale=0.1]{U_t2u_ADF} 
}; 
\node (image) at (9,0.5) { 
  \includegraphics[scale=0.1]{Np_t2u_ADF} 
}; 
\node (image) at (10.5,0.5) { 
  \includegraphics[scale=0.1]{Pu_t2u_ADF} 
};

The relative location of these PNG image ﬁles is directory ../share/ (the naming is for compatibility with our texcollab utility). Instead of adding the ﬁle path in the \includegraphics commands directly, directory ../share/ is speciﬁed in the deﬁnition of \graphicspath in the panel tex ﬁle, so the ﬁles are found by LaTeX without an explicit path speciﬁcation. After compiling, the panel looks like this:

Add some more labels,

\node[right] at (11.5,3.5){\strut \green{\ttwog{}}}; 
\node[right] at (11.5,2){\strut \red{\toneu{}}}; 
\node[right] at (11.5,0.5){\strut \blue{\ttwou{}}}; 
 
\node at (2.5,4.4){\strut U}; 
\node at (3.6,4.4){\strut Np}; 
\node at (4.55,4.4){\strut Pu}; 
 
\node[right] at (7.3,4.4){\strut U}; 
\node[right] at (8.7,4.4){\strut Np}; 
\node[right] at (10.2,4.4){\strut Pu};

and the panel looks like this:

Finally, we remove the grid by setting

% Grid to help placing objects, if so requested 
\renewcommand{\showgrid}{false} % use true or false

and recompile. The panel is now complete. When you create the ﬁgure with the shell script make-panels.sh, the embedded high-resolution images in the resulting PDF get downsampled to 800 DPI in the ghostscript (gs) step, which is usually more than suﬃcient and tends to make the PDF a lot smaller in case there are multiple high-resolution bitmap images included. The xelatex command to create the initial PDF uses the -z3 option, which creates a PDF that is not well compressed but runs much faster than xelatex with default settings.

We sometimes have ten or more such panels for a single publication (incl. SI), and make-panels.sh can be used to compile them all in one go (e.g., to make sure that all compound ﬁgures are up-to-date) without waiting too long. Having downsampled embedded images also helps to speed up scrolling through the manuscript PDF during editing. The ﬁnal version of the panel, shown near the top of this page, is only about 11.5 cm wide without the grid, after it’s been cropped. Most journals would still use a 2-column layout for the ﬁgure.

The PDFs created as described here can be included easily in a LaTeX manuscript, of course. I much prefer this approach over using a gazillion of commands to produce complicated ﬁgures directly within a manuscript’s LaTeX code, although that is of course an option. You can also use the ﬁgures created as described here for inclusion in a document written with a word processor. For this, I usually convert to appropriately sized PNGs with a command such as

f="figure-file-name-without-.pdf-extension"; pdftoppm -r 600 -png ${f}.pdf > ${f}.png

The pdftoppm command is part of the poppler package available for many Linux distributions (package naming varies). Presumably, you can also install it easily in a Linux subsystem on Windows or on a Mac.

Final note: The tikz and pgfplots LaTeX packages are very versatile, so you can use them to draw a lot more complicated things in your ﬁgures than just textual labels. In research, time is of the essence, so I usually end up combining diﬀerent graphics from diﬀerent sources instead of attempting a solution using only LaTeX with tikz and/or pgfplots, say. What I described on this page is a—better, I would argue—replacement for what people usually end up doing, which is placing diﬀerent graphics on a Powerpoint slide, adding labels and other graphical elements, and exporting the slide as a ﬁgure.