I'm still learning this, but here's what I know...

Beyond Nyquist, strange stuff happens.

Nyquist frequency is 1/2 of the sample rate. In theory, you can't do ANYTHING in the digital realm above Nyquist, because it is the maximum frequency you can reproduce at a given sample rate, i.e., when each sample is +1,-1,+1,-1..., it takes 44.1K samples to represent a 22.05K frequency.

Also consider the possibility of a 22.05K frequency, where the sample just happens to land on every 0 crossing. In this case, the sound is perfectly 'invisible' to the digital realm. And in truth, you are just as likely to experience this as you are to get a perfect +1,-1,+1,-1... series. Thus, very strange things happen even as you approach Nyquist. So a 44.1K sampling rate is actually fairly useless at representing signals above 20K. Fortunately, 20K is beyond the range we can hear.

So what happens when you try to represent a high frequency signal that CANNOT be represented digitally because there aren't enough "plotting points"? Harmonics at best, noise at worst.

As to how this can make 1K look like 44101K, take a peek at the first chart under "Sampling Sinusoidal Functions" at the following link (two different sinusoids that fit the same set of samples):

http://en.wikipedia.org/wiki/Aliasing