WARNING - LONG TEDIOUS POST AHEAD.
The problem with answering questions like this definively is lack of data. The relevant variables can probably be decomposed into three questions:
1. What frequency range does the target (music) occupy
2. What frequency range can the human ear/brain hear over
3. Given that playback systems must distort (i.e., are not faithful to the orginal) the signal somewhat, what type of distortions are harmful and what type are inaudible or at least benign
A major problem with answering this question is that (2) is not an engineering question and is very difficult to determine. It requires complex psychophysical experiments - this is basically the scientific quantification of subjective experience.
The coventional wisdom is that the structure of the ear does not allow the transduction of soundawaves below about 20-30Hz, although we can feel low bass through vibratio of the chest cavity and the vestibular organs. Careful studies (and subjective impression) are pretty convincing that reproducing very low bass, while not heard, adds to the realism, enjoyment etc of music and movies. Seveeral thousand published scientific studies speak to the issues so flippantly described in this paragraph.
The conventional wisdom is also that we can't hear much above 20Khz (or much less). This is based on the observations: people do not report hearing tones higher than this, that the basilar membrane does not have resonant points for these frequencies, and that there is little (no?) auditory brainstem activites for tones at higher frequencies, to name but a few reasons. In fact for frequencies above 5kHz people have a sense the pitch is "high" but do not form a subjective impression of a "note" (like C#, F, etc) and melodies transposed to this range do not sound musical. It is instructive to note that the fundamental of the highest piano note is 4186Hz, that is around 4k - well inside this "I can hear a defnite note" area. However, while the fundamental pitch of instruments rarely goes above 5k there are harmonics in most instruments that go out to 30K or even much higher. Remember however, that while there are 8-9 octaves in the muscial range, teh 20-40K band is only one more octave and the next octave is 80K and almost no-one is willing to say there is anything useful out there.
The assumption underlying a lot of gear design (CD players particularly) is that since there is no measureable human response to sounds above 20K cutting these frequencies out will have no impact on the perception of recorded music. We can save an engineering headache by not bothering to deal with sounds above 20K. Again, the CONVENTIONAL wisdom is that we don't need to record this, and while there are thousands of articles on ~20K being the limit, there are only a few dozen articles which show any sort of subjective or neural response to >20K frequencies, and at least half of these have methodological flaws.
Three tentative possbilities emerge
i) we are wrong about the limits of human hearing and need to build much wider bandwidth gear (assuming there is anything meaningfully musical up beyond 20K)
ii) we are right about the limits of human hearing, which means that any shortcomings are the result of (3) distortions in the playback medium
<science bit over, the rest just my opinion>
I have gone on record as being a fan (but willing to change my mind) of (ii). Filtering (in the form of deliberate or unintended band limiting) introduces artifacts into the signal, and I believe it is these artifacts which harm the sound. Wide-bandwidth sources minimise these filtering artifacts and lead to better reproduction of in-band sounds.
My 2c again.
Very interesting discussion. Thanks to everyone who is chiming in.
T.
