The first issue is the usage of the staircase kerning boxes of
opentype math fonts. They are obviously not applied in expressions
like $f_j$ using LuaLaTeX. The index is rendered too far away from
the glyph. If we use XeLaTeX we get the correct output.
It works okay with luatex 1.09 (miktex or dev-TeXLive), but fails in a
current texlive 2018.
Okay, that is good news. How can I receive the required luatex file? I am on a Mac and TeXLive is the standard distribution there. Till now I did not know anything of dev-TeXLive and how to install it. Would it be possible to simply send the current luatex 1.09 file to me so that I can replace the file in the standard distribution? Or would that require more complicated actions?
The second issue is present in both LuaLaTeX and XeLaTeX. If we
have an index together with an exponent like j the index is rendered
too high – in fact it is risen compared to the expression that uses
an index only. A typical example is $f_n^j$.
The sup/sub pair is higher placed than with context. You could move it
down (at least with luatex) with
It is obvious that the higher placement in LuaLaTeX and XeLaTeX cannot be intended and thus should be considered as a bug.
Have a look at the attached file. This time I have chosen Latin Modern Math because it is a little bit more significant there. Notice that $f_n$ and even $f_n^m$ have the correct and standard placement while $f_n^j$ shows the higher placement of the index $n$. There is no reason why the $n$ should be placed higher if the exponent is $j$ instead of $m$. On the contrary it would make much more sense if the $n$ would be placed lower(!) and not higher if the exponent goes below the baseline and has a descender height > 0.
\documentclass{article}
\usepackage{unicode-math}
\setmathfont{Cambria Math}
\begin{document}
$f_n^j$
\Umathsupsubbottommax\textstyle =2pt
$f_n^j$
\end{document}
unicode-math could probably offer some option to set this.
Maybe this could be a workaround as long as the bug persists. Has anyone an idea how this could be done as standard procedure using unicode-math?