... it's not evident to me why the need for null concatenation ...

I haven't looked at the docs or source at all, but apparently the  selectElementsByContent() method does not like to be passed a reference. The  qr// operator returns a  Regexp reference. In essence, the  "". qr/faith/i expression replicates the approach of choroba above by stringizing this reference before passing it as an argument, but still allowing all the facility of the  qr// operator to form the pattern object in the first place: all the i-s and t-s get properly dotted and crossed, respectively. (You crotted an i in the solution you attempted here.) I feel it's almost always better to use  qr// to build patterns than than to try to express the patterns as strings using string operators that are almost, but not quite, the same.

Here's an example of conversion of a  Regexp object to a string and its subsequent use as a string in a match.

c:\@Work\Perl\monks>perl -wMstrict -le
"my $rx = qr{ (?i) f a i t h }xms;
 print 'qr// object: ', ref_or_not($rx);
 print $rx;
 ;;
 my $rx_stringized = '' . $rx;
 print 'stringization: ', ref_or_not($rx_stringized);
 print qq{'$rx_stringized'};
 ;;
 my $s = 'UnFaItHfUl';
 $s =~ $rx_stringized;
 print qq{'$&'};
 ;;
 sub ref_or_not {
   my $scalar = shift;
   return ref $scalar ? ref $scalar : 'not a ref';
   }
"
qr// object: Regexp
(?^msx: (?i) f a i t h )
stringization: not a ref
'(?^msx: (?i) f a i t h )'
'FaItH'
[download]

Note that the  =~ operator is quite happy to use a pure string as a match pattern.