(unicode-math or newtxmath) + \newcommand{\nmodels}{\not\models} + $\not\models$ = stall

Question

Feeding the input

\RequirePackage{ifthen}
\RequirePackage{ifxetex,ifluatex}
\newif\ifxetexorluatex
\ifxetex
  \xetexorluatextrue
\else
  \ifluatex
    \xetexorluatextrue
  \else
    \xetexorluatexfalse
  \fi
\fi
\documentclass{standalone}
\ifxetexorluatex
\usepackage{unicode-math}
\else
\usepackage{newtxmath}
\fi
\newcommand{\nmodels}{\not\models}%%% or \providecommand, or \providecommand* instead of \newcommand; the effect is the same.
\begin{document}
\(\nmodels\) %%% or \(\not\models\); the effect is the same.
\end{document}

to any of pdflatex, xelatex, or lualatex causes an infinite loop. Well, in case of XeLaTeX + unicode-math enters infinite loop if \nin is defined? you can use \notin instead of \not\in, but what do you do if you like the shape of \models (rather than that of \vDash), want a negated form of it, and want to call it \nmodels? The form of \nvDash is different, so, it's not advisable to use both \models and \nvDash in the same document. Using (the whole of) MnSymbols is out of question: it may still have bugs (I knew it had some 10 years ago and stopped using it then), changes quite a a few symbols, and clashes with NewTX fonts.

Answer 1

The current implementation of \not does essentially the same in pdflatex as in (Xe|Lua)LaTeX with unicode-math.

More precisely, \not takes the next token, say \foo and first checks whether \notfoo is defined; if the test succeeds, \notfoo is used. Otherwise the next check is whether \nfoo is defined. Again, if this test succeeds, \nfoo is used. Otherwise LaTeX does \n@tch@r\foo.

Now you see why an infinite loop is started: \nmodels does \not\models; since \nmodels is defined, it is used, which does \not\models…

You can avoid the issue by applying \not to \relax (hoping that neither \notrelax nor \nrelax are defined, which is unlikely).

\documentclass{standalone}
\usepackage{iftex}
\iftutex
  \usepackage{unicode-math}
\else
  \usepackage{newtxmath}
\fi

\newcommand{\nmodels}{\not\relax\models}%

\begin{document}

\(\nmodels\)

\(\not\models\)

\end{document}

Anyway, this is not a particularly good way to negate \models. Compare with the following.

\documentclass[border=4]{standalone}
\usepackage{iftex}
\iftutex
  \usepackage{unicode-math}
\else
  \usepackage{newtxmath}
\fi
\usepackage{centernot}

\newcommand{\nmodels}{\centernot\models}%

\begin{document}

\(\nmodels\)

\(\not\models\)

\end{document}

Answer 2

The definition of \not is getting too clever by half by looking up whether \nmodels or \notmodels already exist.

It works if you change the name:

\documentclass{standalone}
\usepackage{iftex}

\iftutex
  \usepackage{unicode-math}
\else
  \usepackage{newtxmath}
\fi

%\newcommand{\aintmodel}%
  %{\mathrel{\ooalign{\(\models\)\cr\hidewidth\(\mathslash\)\hidewidth}}}

\newcommand\aintmodel{\not\models}

\begin{document}
\( \aintmodel \)
\end{document}

This lookup gets performed whenever the symbol is used, and the definition of \not isn’t fully-expandable, so you can’t define the symbol with \let or \edef and sidestep this.

You could, however, create the symbol with \ooalign instead. (Use / with newtxmath instead of \mathslash.)

Clearly, unicode-math now needs to search for the prefix aint.

Update

You can also call \not without using the name \models, preventing the look-up. For example,

\documentclass{standalone}
\usepackage{iftex}

\usepackage{unicode-math}


\newcommand\nmodels{\mathrel{\not\symbol{"22A7}}}

\begin{document}
\( \nmodels \)
\end{document}