nth root kerning problem with unicode-math font override

Question

I am using both \mathscr and \mathcal to different meanings and expecting different characters to appear.

I recently switched to XeTeX and started using new fonts (TeX Gyre Pagella in this example) with the packages fontspec and unicode-math. As suggested by the answer to this question, I override \mathcal and \mathscr with characters coming from a different font (XITS-math).

Unfortunately this has the effect of mangling the setting of a (seemingly) unrelated math symbol : the nth root.

\documentclass{minimal}
\usepackage{fontspec}
\usepackage{unicode-math}
\setmainfont{TeX Gyre Pagella}
\setmathfont{TeX Gyre Pagella Math}
% Math font overrides ----
\setmathfont[range={\mathcal,\mathbfcal},
             StylisticSet=1,Extension=.otf,BoldFont=*bold]{xits-math}
\setmathfont[range=\mathscr,Extension=.otf,BoldFont=*bold]{xits-math}

\begin{document}
Some math letters:
\[ \mathbb{C} \mathcal{C} \mathscr{C} \mathfrak{C} \]

Here's a big $n$-th root:
\[ F_k(x,y,z)  = \sqrt[n]{\sum_{i=1}^{k} f_i^n} \,. \]
\end{document}

This code produces the following result: All Cs are different, but the n is too close from the radical symbol (√).

Commenting out the math font overrides yield the correct result for the nth root (but \mathcal and \mathscr now produce identical characters):

My questions are then

• Why does overriding \mathscr and \mathcal affects the setting of the nth root ?
• How can I avoid this problem ? (or do I have to change fonts ?)

Answer 1

Edit

Imho with a current unicode-math (2/2018) this is no longer needed.

=====

It looks as if it could work if you reset the fontdimens:

\documentclass{article}
\usepackage{fontspec}
\usepackage{unicode-math}
\setmainfont{TeX Gyre Pagella}
\setmathfont{TeX Gyre Pagella Math}

\ExplSyntaxOn 
 %store the values
\int_step_inline:nnnn {1}{1}{65}{\tl_set:cx{l_tmp_fontdimen_#1_tl}{\the\fontdimen #1  \l__um_font }}
\ExplSyntaxOff

% Math font overrides ----
\setmathfont[range={\mathcal,\mathbfcal},
             StylisticSet=1,Extension=.otf,BoldFont=*bold]{xits-math}
\setmathfont[range=\mathscr,Extension=.otf,BoldFont=*bold]{xits-math}


\ExplSyntaxOn
%restore the values
\int_step_inline:nnnn {1}{1}{65}{\fontdimen #1  \l__um_font =\tl_use:c{ l_tmp_fontdimen_#1_tl}}
\ExplSyntaxOff



\begin{document}
Some math letters:
\[ \mathbb{C} \mathcal{C} \mathscr{C} \mathfrak{C} \]

Here's a big $n$-th root:
\[ F_k(x,y,z)  = \sqrt[n]{\sum_{i=1}^{k} f_i^n} \,. \]
\end{document}

But this is quite experimental ... and only tested with xelatex. (And probably they are still wrong for the other math styles).