On LaTeX-Community.org I've noticed a strange problem with nested \tilde and \breve commands, when amsmath is used.
In this code the X appears also in the output of the second and the third formula:
\documentclass{article} \usepackage{amsmath} \begin{document} \[ \breve{\tilde{X}} \] \[ \breve{\breve{{T}}\breve{{v}}} \] \[ \tilde{{\tilde{T}}{\tilde{v}}} \] \end{document} 
If I don't load amsmath, the code works fine. This seems to be a bug in amsmath. How can it be solved?
The problem comes up already with a "more minimal" example:
\documentclass{article} \usepackage{amsmath} \begin{document} $\breve{\breve{a}\breve{b}}$ \end{document} which produces
! Undefined control sequence. \macc@adjust ->\dimen@ \macc@kerna \advance \dimen@ \macc@kernb \kern -\dimen@ thus showing that the problem is not only in the repetition of the symbol. It happens with all math accents defined in terms of \mathaccentV,
\hat \check \tilde \acute \grave \dot \ddot \bar \vec \mathring because these accents work by looking if their argument contains another accented symbol, in order to stack precisely the accents. This requires doing some global definitions, but somehow, if the argument contains two of these accents, the mechanism fails.
The "repeated symbol" is contained in the macro \macc@nucleus: after $\tilde{\breve{X}$, it expands to X and this is why this symbol is repeated (but I've not digged much into the details), since the definition of \macc@nucleus is done via \gdef.
The amsmath documentation doesn't point out that only stacked accents on one symbols should be used and, actually, it's safe to put an accent on a subformula provided it doesn't contain accents.
As the macros are quite complex and require global assignments, it seems quite difficult to do surgery on them, so a different approach is easier. Define
\newsavebox{\accentbox} \newcommand{\compositeaccents}[2]{% \sbox\accentbox{$#2$}#1{\usebox\accentbox}} Now \breve{\breve{a}\breve{b}} can be changed into
\compositeaccents{\breve}{\breve{a}\breve{b}} and all goes well. As one can see, the argument is typeset before applying the "global accent" and stored in a bin, over which we can safely put the "global accent".
Should one need such buildups also in superscripts or subscripts
\newcommand{\compositeaccentsX}[2]{% \let\accenttemp#1\mathpalette\docompositeaccents{#2}} \def\docompositeaccents#1#2{\compositeaccents\accenttemp{#1#2}} and $A_{\compositeaccentsX{\breve}{\breve{a}\breve{b}}}$ will work. It's better to stick with the simpler command, as the "extended" one requires to typeset four times the same formula.
Curiously enough, the accents package shows a bug in situations like this one:
\documentclass{article} \usepackage{amsmath,accents} \begin{document} $\breve{\breve{a}\breve{b}}$ \end{document} will give no error, but will eat up the "a".
Very nice 10000th question on TeX.SE!
this is certainly a bug, and has already been reported as such to [email protected]. it has been added to the list for the next overhaul of the amsmath package.
here are two additional small tests that provide useful information:
\documentclass[11pt]{article} \usepackage[]{amsmath} \begin{document} \[ \breve{\breve{T} v} \] \end{document} this results in everything being terribly misplaced.
\documentclass[11pt]{article} \usepackage[]{amsmath} \begin{document} \[ \breve{\breve{T} \breve{v}} \] \end{document} this crashes with
! Undefined control sequence. \macc@adjust ->\dimen@ \macc@kerna \advance \dimen@ \macc@kernb \kern -\dimen@ l.6 \breve{\breve{T} \breve{v}} thanks for the additional diagnosis and suggestions.
If you load the accents package (after loading amsmath) with the single option set, you won't get the problems you describe:
\documentclass{article} \usepackage{amsmath} \usepackage[single]{accents} \begin{document} $\breve{\breve{a}\breve{c}}$, $\tilde{\tilde{T}\tilde{v}}$ \end{document} 
Observe that the positioning of the upper (second-storey) accent will be quite poor if the accented glyphs below it have significantly different heights and/or widths (as is the case with the second example above). If they're available, one should use the wide variants of the accent symbols for the upper (second-story) accents, e.g.,
$\widetilde{\tilde{T}\tilde{v}}$ Of course, there is (to my knowledge) no "wide" version of the breve accent, so this piece of advice can't be applied for all accents...
Addendum: A caveat to the generality of this solution: I've just discovered that if one uses the accents package for some of the additional accents it provides (such as \ddddot), you can't use these additional accents in the upper position in the examples above as, in that case, the first letter (but, curiously, not its accent) will be gobbled up. I've only tested this with the ddddot accent, to be sure, but this problem may be wider spread. :-(
Another approach could be resetting \macc@nucleus at a suitable place, such as here:
\documentclass{article} \usepackage{amsmath} \let\amsmathaccentV\mathaccentV \makeatletter \def\mathaccentV#1#2#3#4#5{% \amsmathaccentV{#1}{#2}{#3}{#4}{#5}% \global\let\macc@nucleus\@empty } \makeatother \begin{document} \[ \breve{\tilde{X}} \] \[ \breve{\breve{{T}}\breve{{v}}} \] \[ \tilde{{\tilde{T}}{\tilde{v}}} \] \end{document} Which gives:
However, there would still be the open problem that the complete accented expression might not have its natural width, which can be seen if we let characters or symbols follow.
Mentioning this on the way of finding a good fix for amsmath.
\macc@kerna still occurs in that MWE. Just adding a thought which may become a step in fixing. At least it fixes the question.
\o/\breve{\tilde{X}} \vphantom{\breve{}}