Accented characters in LaTeX can be produced using commands such as
\"a etc. The precise effect of such commands depends on the font
encoding being used. When using a font encoding that contains the
accented characters as individual glyphs (such as the T1 encoding,
in the case of \"a) words that contain such accented characters can
be automatically hyphenated. For font encodings that do not contain
the requested individual glyph (such as the OT1 encoding) such a
command invokes typesetting instructions that produce the accented
character as a combination of character glyphs and diacritical marks
in the font. In most cases this involves a call to the TeX
primitive \accent. Glyphs constructed as composites in this way
inhibit hyphenation of the current word; this is one reason why the
T1 encoding is preferable to the original TeX font encoding
OT1.
It is important to understand that commands like \"a in LaTeX2e
represent just a name for a single glyph (in this case `umlaut a') and
contain no information about how to typeset that glyph--thus it does
not mean `put two dots on top of the character a'. The
decision as to what typesetting routine to use will depend on the
encoding of the current font and so this decision is taken at the last
minute. Indeed, it is possible that the same input will be typeset in
more than one way in the same document; for example, text in section
headings may also appear in table of contents and in running heads; and
each of these may use a font with a different encoding.
For this reason the notation \"a is not equivalent to:
\newcommand \chara {a} \"\chara
In the latter case, LaTeX does not expand the macro \chara but
simply compares the notation (the string \"\chara) to its list of
known composite notations in the current encoding; when it fails to
find \"\chara it does the best it can and invokes the typesetting
instructions that put the umlaut accent on top of the expansion of
\chara. Thus, even if the font actually contains `ä' as an
individual glyph, it will not be used.
The low-level accent commands in LaTeX are defined in such a way
that it is possible to combine a diacritical mark from one font with a
glyph from another font; for example, \"\textparagraph will produce
¨ 
. The umlaut here
is taken from the OT1 encoded font cmr10 whilst the paragraph sign
is from the OMS encoded font cmsy10. (This example may be
typographically silly but better ones would involve font encodings
like OT2 (Cyrillic) that might not be available at every
site.)
There are, however, restrictions on the font-changing commands that
will work within the argument to such an accent command. These are
TeXnical in the sense that they follow from the way that TeX's
\accent primitive works, allowing only a special class of commands
between the accent and the accented character.
The following are examples of commands that will not work correctly as
the accent will appear above a space: the font commands with text
arguments (\textbf{...} and friends); all the font size declarations
(\fontsize and \Large, etc.); \usefont and declarations that
depend on it, such as \normalfont; box commands (e.g. \mbox{...}).
The lower-level font declarations that set the attributes family,
series and shape (such as \fontshape{sl}\selectfont) will produce
correct typesetting, as will the default declarations such as
\bfseries.