Nyquist theorem states that a non-variant signal freqency can be reproduced
that is 1/2 the sample rate. Unfortunately, music that is invariant is not
terribly interesting. Thus, the common wisdom that 44.1KHz sampling can
reproduce 22 KHz music is not true.

A seminal paper from MIT shows that distortion related to sampling must
consider both the sample rate and the target word size. For today's
CDs--that is 16 bits. Thus, according to this paper, a minimum of 8X
frequency is required--10X is better. Working backwards, that means that CD
technology can only reproduce, at best, 5.5 KHz before distortion starts to
enter in. This is independant of the construction of filters and assumes a
boxcar filter (impossible in real life.)

Other solutions have worked hard to reduce this problem by oversampling,
adding bits, etc. All these solutions smooth the distortion created by the
original system, but they can not add information back in that is lost.
What they can do is create better sounding music by smoothing out the
jaggies in the distortion.