Last August, the MathJax Consortium quietly released version 3.0.0 of MathJax, a complete rewrite of MathJax from the ground up with the goal of modernizing MathJax’s internal infrastructure, bringing it more flexibility for use with contemporary web technologies, making it easier to use with NodeJS for pre-processing and server-side support, and making it faster to render your mathematics.

Because this was a major update to MathJax, with a completely new application programmer interface (API), and because not all the MathJax version 2 features were ported to version 3, we wanted to wait a bit before making a formal announcement of the new version. During the fall, we collected bug reports and responded to requests for code examples, and in early February 2020, we released version 3.0.1, a bug-fix version that resolved a number of important issues with 3.0.0.

Today, we release version 3.0.5, which fixes additional bugs, and puts MathJax v3 into a stable enough position to warrant formal announcement of the availability of version 3. Note that versions 3.0.2 and 3.0.3 had problems with their `npm`

deployment, and have been removed. Version 3.0.4 is the bug-fix release that fixes most issues that have been reported in v3.0.1, and v3.0.5 is a hot-fix release to resolve an important problem with v3.0.4.

We are pleased to present this result of over three years of work by the MathJax developers. The following material comes from the What’s New in v3 section of the MathJax documentation.

For information about converting from version 2 to version 3, see the section on Upgrading from v2. In particular, note that there is a configuration conversion tool to help you move your version 2 configuration to the new version 3 format. There are a number of examples for using MathJax in web pages in the MathJax Web Demos repository, and examples of using MathJax in NodeJS applications in the MathJax Node Demos repository.

# Features of Version 3

## Improved Speed

There were a number of design goals to the version 3 rewrite. A primary one was to improve the rendering speed of MathJax, and we feel we have accomplished that. Because the two versions operate so differently, it is difficult to make precise comparisons, but in tests that render a complete page with several hundred expressions, we see a reduction in rendering time of between 60 and 80 percent, depending on the browser and type of computer.

## More Flexibility

Another goal was to make MathJax 3 more flexible for web developers using MathJax as part of a larger framework, while still keeping it easy to use in simple settings. To that end, we have broken down the actions that MathJax takes into smaller units than in version 2, and made it possible to call on them individually, or replace them with alternative versions of your own. For example, the typesetting process has been broken into a number of pieces, including finding the math in the page, compiling it into the internal format (MathML), getting metric data for the location of the math, converting the math into the output format, inserting it into the page, adding menu event handlers, and so on. You have control over which of these to perform, and can modify or remove the existing actions, or add new ones of your own. See the `renderActions`

documentation for details.

## Improved Support for Assistive Technology

Accessibility of mathematics has always been an important focus of MathJax, and that continues to be the case in version 3. The speech-rule engine (SRE) that underlies most of MathJax’s assistive support has been improved for version 3 to include more sophisticated speech generation, support for Nemeth Braille output, localization in several languages, and more user control over the speech that is generated. See the Accessibility Features page for more details.

## Synchronous Conversion

A key feature that we wanted to include in version 3 is the ability to run MathJax synchronously, and in particular, to provide a function that can translate an input string (say a TeX expression) into an output DOM tree (say an SVG image). This was not really possible in version 2, since its operation was inherently asynchronous at a fundamental level. With MathJax version 3, this is straight-forward, as we provide a synchronous typesetting path, both within the page, and for individual expressions, provided you load all the components you need ahead of time. See Typesetting and Converting Mathematics for details.

## No Queues, Signals, Callbacks

One of the more difficult aspects of working with MathJax version 2 was having to synchronize your actions with those of MathJax. This involved using *queues*, *callbacks*, and *signals* to mediate the asynchronous actions of MathJax. Since these were not standard javascript paradigms, they caused confusion (and headaches) for many developers trying to use MathJax. With version 3, MathJax has the option of working synchronously (as described above), but it still allows for asynchronous operation (*e.g.*, to allow TeX’s `\require`

command to load extensions dynamically) if you wish. This no longer relies on queues, callbacks, and signals, however. Instead, these actions are managed through the ES6 promise, which is a javascript standard and should make integrating MathJax into your own applications more straight-forward.

## Package Manager Support

Because MathJax version 2 used its own loading mechanism for accessing its components, and because there was no method for combining all the pieces needed by MathJax into one file, MathJax did not work well with javascript packaging systems like `webpack`

. Version 3 resolves that problem, so it should interoperate better with modern web workflows. You can make your own custom single-file builds of MathJax (see Making a Custom Build of MathJax), or can include it as one component of a larger asset file.

## MathJax Components

MathJax 3 still provides a loading mechanism similar to the one from version 2, however, so you can customize the extensions that it loads, meaning that you only load the ones you need (though this does require that you use MathJax in its asynchronous mode). The various pieces of MathJax have been packaged into “components” that can be mixed and matched as needed, and which you configure through a global `MathJax`

variable (see Examples in a Browser). This is how MathJax is being distributed through the various CDNs that host it. When loaded this way, MathJax will automatically set up all the objects and functions that you need to use the components you have loaded, giving you easy access to typesetting and conversion functions for the input and output formats you have selected. See the section on The MathJax Components for more information. You can also create your own custom components to complement or replace the ones provided on the CDN (see A Custom Extension for more).

## Startup Actions

If you use any of the combined component files, MathJax will perform a number of actions during its startup process. In particular, it will create the input and output jax, math document, DOM adaptor, and other objects that are needed in order to perform typesetting in your document. You can access these through the `MathJax.startup`

object, if you need to. MathJax will also set up functions that perform typesetting for you, and conversion between the various input and output formats that you have loaded. This should make it easy to perform the most important actions available in MathJax. See Typesetting and Converting Mathematics for more details.

## Server-Side MathJax

While MathJax 2 was designed for use in a web browser, an important use case that this left unaddressed is pre-processing mathematics on a server. For version 2, we provided mathjax-node to fill this gap, but it is not as flexible or easy to use as many would have liked. MathJax 3 resolves this problem by being designed to work with node applications in essentially the same way as in a browser. That is, you can load MathJax components, configure them through the MathJax global variable, and call the same functions for typesetting and conversion as you do within a browser. This makes parallel development for both the browser and server much easier.

Moreover, node applications can access MathJax modules directly (without the packaging used for MathJax components). This gives you the most direct access to MathJax’s features, and the most flexibility in modifying MathJax’s actions. See Examples of MathJax in Node for examples of how this is done.

## ES6 and Typescript

MathJax 3 is written using ES6 modules and the Typescript language. This means the source code includes type information (which improves the code reliability), and allows MathJax to be down-compiled to ES5 for older browsers while still taking advantage of modern javascript programming techniques. It also means that you can produce pure ES6 versions of MathJax (rather than ES5) if you wish; these should be smaller and faster than their ES5 equivalents, though they will only run in modern browsers that support ES6, and so limit your readership. We may provide both ES6 and ES5 versions on the CDN in the future.

# New Features for Existing Components

In addition to the new structure for MathJax described above, some new features have been added to existing pieces of MathJax.

## TeX Input Extensions

There are two new TeX input extensions: braket and physics. Also, some functionality that was built into the TeX input jax in version 2 has been moved into extensions in version 3. This includes the macros configuration option, the tag formatting configuration options, and the require macro. The new autoload extension replaces the older autoload-all extension, is more configurable, and is included in the TeX input components by default. There a several extensions that are not yet ported to version 3, including the `autobold`

, `mediawiki-texvc`

, and the various third-party extensions.

## SVG Output

The SVG output for equations with labels has been improved so that the positions of the labels now react to changes in the container width (just like they do in the HTML output formats).

## Improved Expression Explorer

The interactive expression explorer has been improved in a number of ways. It now includes better heuristics for creating the speech text for the expressions you explore, provides more keyboard control of the features in play during your exploration, adds support for braille output, adds support for zooming on subexpressions, and more. See the Accessibility Features page for more details.