I would like to add to this that real instruments are much more complicated in their sound. If you take a look at the frequency histogram of one of those, you'll realize that what you might call a theory of pure frequencies, while perhaps adequate for explaining melody (several notes put together) is nowhere enough for explaining sound. Just like salty/fatty/sweet/sour is nowhere enough for explaining taste.
This is also why it is so incredibly difficult to make a realistic sounding synthesizer.
If you want to play with this, search for a sample pack and try running a couple of the sounds from that through sox to get a histogram.
I'm not saying this is wrong, but the Nyquist-Shannon sampling theorem states that a "theory of pure frequencies" is sufficient to reconstruct a band-limited signal from samples. What's the missing element?
I'd hasard a guess that the problem is not that you can't perfectly reproduce a sound of an instrument being played. However when you simulate it using samples you reduce the sound to a sum of independent recordings, whereas a real instrument will behave differently when you play multiple notes because of the material.
1. (as you say) How playing multiple notes at the same time changes the way the instrument responds
2. How playing at different intensities changes how the instrument responds.
3. The many subtle ways your physical interaction with the instrument change the sound
4. The way playing notes in succession at different rates can alter the sound
5. The physical space or choice of amplification can affect the instrument - even an acoustic guitar can "feed back" on itself to some degree
6. A bunch of other things I haven't thought of.
Sophisticated samplers (and sophisticated sample libraries) can simulate some of the above. But physical modelling synths are probably a better way forward.
This is also why it is so incredibly difficult to make a realistic sounding synthesizer.
If you want to play with this, search for a sample pack and try running a couple of the sounds from that through sox to get a histogram.