Hans Petter Langtangen, Simula Research Laboratory and University of Oslo
Date: May 29, 2016
Summary. The purpose of this document is to test LaTeX math in DocOnce with various output formats. Most LaTeX math constructions are renedered correctly by MathJax in plain HTML, but some combinations of constructions may fail. Unfortunately, only a subset of what works in html MathJax also works in sphinx MathJax. The same is true for markdown MathJax expresions (e.g., Jupyter notebooks). Tests and examples are provided to illustrate what may go wrong.
The recommendation for writing math that translates to MathJax in
html, sphinx, and markdown is to stick to the environments \[
... \]
, equation
, equation*
, align
, align*
, alignat
, and
alignat*
only. Test the math with sphinx output; if it works in that
format, it should work elsewhere too.
The current version of the document is translated from DocOnce source to the format ipynb.
We can get an inline equation
$u(t)=e^{-at}$
rendered as $u(t)=e^{-at}$.
An equation with number,
!bt
\begin{equation} u(t)=e^{-at} \label{eq1a}\end{equation}
!et
looks like
Maybe this multi-line version is what we actually prefer to write:
!bt
\begin{equation}
u(t)=e^{-at}
\label{eq1b}
\end{equation}
!et
The result is the same:
We can refer to this equation through its label eq1b
: (2).
MathJax has historically had some problems with rendering many LaTeX
math environments, but the align*
and align
environments have
always worked.
!bt
\begin{align*}
u(t)&=e^{-at}\\
v(t) - 1 &= \frac{du}{dt}
\end{align*}
!et
Result:
!bt
\begin{align}
u(t)&=e^{-at}
\label{eq2b}\\
v(t) - 1 &= \frac{du}{dt}
\label{eq3b}
\end{align}
!et
Result:
We can refer to the last equations as the system (3)-(4).
In LaTeX, equations within an align
environment is automatically
given numbers. To ensure that an html document with MathJax gets the
same equation numbers as its latex/pdflatex companion, DocOnce
generates labels in equations where there is no label prescribed. For
example,
!bt
\begin{align}
u(t)&=e^{-at}
\\
v(t) - 1 &= \frac{du}{dt}
\end{align}
!et
is edited to something like
!bt
\begin{align}
u(t)&=e^{-at}
\label{_auto5}\\
v(t) - 1 &= \frac{du}{dt}
\label{_auto6}
\end{align}
!et
and the output gets the two equation numbered.
!bt
\begin{align}
\frac{\partial u}{\partial t} &= \nabla^2 u, & x\in (0,L),
\ t\in (0,T]\\
u(0,t) &= u_0(x), & x\in [0,L]
\end{align}
!et
The result in ipynb becomes
A better solution is usually to use an alignat
environment:
!bt
\begin{alignat}{2}
\frac{\partial u}{\partial t} &= \nabla^2 u, & x\in (0,L),
\ t\in (0,T]\\
u(0,t) &= u_0(x), & x\in [0,L]
\end{alignat}
!et
with the rendered result
align/alignat environments with equation numbers are anti-aligned!
In the sphinx
, ipynb
, and pandoc
output formats, DocOnce rewrites
the equations in an alignat
environment as individual equations in
equation
environments (or more precisely, sphinx
can work with
alignat*
so only numbered alignat
equations get rewritten as individual
equations). If the alignment is somewhat important, try the best with a
manual rewrite in terms of separate equation
environments, and stick to
align*
and alignat*
in sphinx
.
If DocOnce had not rewritten the above equations, they would be rendered in ipynb as
$$ \begin{alignat}{2} \frac{\partial u}{\partial t} &= \nabla^2 u, & x\in (0,L), \ t\in (0,T]\\ u(0,t) &= u_0(x), & x\in [0,L] \end{alignat} $$Let us try the old eqnarray*
environment.
!bt
\begin{eqnarray*}
u(t)&=& e^{-at}\\
v(t) - 1 &=& \frac{du}{dt}
\end{eqnarray*}
!et
which results in
!bt
\begin{eqnarray}
u(t)&=& e^{-at}
\label{eq2c}\\
v(t) - 1 &=& \frac{du}{dt}
\label{eq3c}
\end{eqnarray}
!et
which results in
!bt
\begin{multline}
\int_a^b f(x)dx = \sum_{j=0}^{n} \frac{1}{2} h(f(a+jh) +
f(a+(j+1)h)) \\
=\frac{h}{2}f(a) + \frac{h}{2}f(b) + \sum_{j=1}^n f(a+jh)
\label{multiline:eq1}
\end{multline}
!et
gets rendered as
and we can hopefully refer to the Trapezoidal rule as the formula (13).
Although align
can be used to split too long equations, a more obvious
command is split
:
!bt
\begin{equation}
\begin{split}
\int_a^b f(x)dx = \sum_{j=0}^{n} \frac{1}{2} h(f(a+jh) +
f(a+(j+1)h)) \\
=\frac{h}{2}f(a) + \frac{h}{2}f(b) + \sum_{j=1}^n f(a+jh)
\end{split}
\end{equation}
!et
The result becomes
!bt
\[ \frac{\partial\u}{\partial t} +
\u\cdot\nabla\u = \nu\nabla^2\u -
\frac{1}{\varrho}\nabla p,\]
!et
and the inline expression $\nabla\pmb{u} (\pmb{x})\cdot\pmb{n}$
(with suitable newcommands using pmb)
get rendered as
and $\nabla\pmb{u} (\pmb{x})\cdot\pmb{n}$.
Somewhat nicer fonts may appear with the more modern \bm
command:
!bt
\[ \frac{\partial\ubm}{\partial t} +
\ubm\cdot\nabla\ubm = \nu\nabla^2\ubm -
\frac{1}{\varrho}\nabla p,\]
!et
(backslash ubm
is a newcommand for bold math $u$), for which we get
Moreover,
$\nabla\boldsymbol{u}(\boldsymbol{x})\cdot\boldsymbol{n}$
becomes $\nabla\boldsymbol{u}(\boldsymbol{x})\cdot\boldsymbol{n}$.
Warning.
Note: for the ipynb format, \bm
was substituted by DocOnce
to \boldsymbol
.
Finally, we collect some problematic formulas in MathJax. They all work fine in LaTeX. Most of them look fine in html too, but some fail in sphinx, ipynb, or markdown.
The LaTeX code
!bt
\[ {\color{blue}\frac{\partial\u}{\partial t}} +
\nabla\cdot\nabla\u = \nu\nabla^2\u -
\frac{1}{\varrho}\nabla p,\]
!et
results in
but correct rendering in ipynb requires omitting the \color
command:
!bt
\[ \bar\f = f_c^{-1}\f,\]
!et
which for ipynb output results in
With sphinx, this formula is not rendered. However, using curly braces for the bar,
!bt
\[ \bar{\f} = f_c^{-1}\f,\]
!et
makes the output correct also for sphinx:
!bt
\begin{align}
\begin{pmatrix}
G_2 + G_3 & -G_3 & -G_2 & 0 \\
-G_3 & G_3 + G_4 & 0 & -G_4 \\
-G_2 & 0 & G_1 + G_2 & 0 \\
0 & -G_4 & 0 & G_4
\end{pmatrix}
&=
\begin{pmatrix}
v_1 \\
v_2 \\
v_3 \\
v_4
\end{pmatrix}
+ \cdots
\label{mymatrixeq}\\
\begin{pmatrix}
C_5 + C_6 & -C_6 & 0 & 0 \\
-C_6 & C_6 & 0 & 0 \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0
\end{pmatrix}
\frac{d}{dt} &=
\begin{pmatrix}
v_1 \\
v_2 \\
v_3 \\
v_4
\end{pmatrix} =
\begin{pmatrix}
0 \\
0 \\
0 \\
-i_0
\end{pmatrix}
\end{align}
!et
which becomes
The first equation do not render correctly in ipynb.
The same matrices without labels in an align*
environment:
!bt
\begin{align*}
\begin{pmatrix}
G_2 + G_3 & -G_3 & -G_2 & 0 \\
-G_3 & G_3 + G_4 & 0 & -G_4 \\
-G_2 & 0 & G_1 + G_2 & 0 \\
0 & -G_4 & 0 & G_4
\end{pmatrix}
&=
\begin{pmatrix}
v_1 \\
v_2 \\
v_3 \\
v_4
\end{pmatrix}
+ \cdots \\
\begin{pmatrix}
C_5 + C_6 & -C_6 & 0 & 0 \\
-C_6 & C_6 & 0 & 0 \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0
\end{pmatrix}
\frac{d}{dt} &=
\begin{pmatrix}
v_1 \\
v_2 \\
v_3 \\
v_4
\end{pmatrix} =
\begin{pmatrix}
0 \\
0 \\
0 \\
-i_0
\end{pmatrix}
\end{align*}
!et
The rendered result becomes