Sampling is a term that you will see everywhere in regard to computer imaging. Sampling is a technique used to examine a particular element within a dataset (scene) at a given time. You can think of sampling as a large buffet table. At the buffet, you are free to take whatever you wish, whenever you want (just don’t forget to pay.) In a similar fashion, a render engine can look at the buffet of your scene, and take samples to determine the appearance of the final rendered image or animation. And, you get to tell the render engine what kind of samples you want to use, when you want to use them, and how many samples and under what conditions you wish to make. Sampling is your friend. Get to know how it works, and you will be able to make great renderings in the shortest amount of time, every time. That’s the good news.
The bad news is that you need to learn this stuff, and it can seem like a jargon-fest or aplhabet soup at times. Just take it slow and easy, and you’ll get the hang of it. It happens to all of us.
Sampling is a necessary evil in anything digital. You may be familiar with music quality sampling, such as 44Khz, or 128 bit, etc. The old analog recording methods, such as magnetic tape and film, can lay down a continuous waveform of sound and light. The fidelity of those original recordings are limited by the quality of the equipment used, the abilities of the person making the recording, and the constraints of the magnetic tape or film recording media. There were always thresholds associated with those methods, but they tended to be more natural than digital (or perhaps that was just our perception, since those methods have been with us for over 100 years now.)
Digital sampling can be done in a continuous manner like analogue. Instead, it must be sampled in many little slices of discrete bits, strung together and then played back or viewed as one cohesive event. If you were to draw an analogue wave, you would see a nice clean curve. If you were to plot the digitial version of the same wave, you would see something that looks like a bar graph. In fact, that’s what it is. Now imagine the bar graph conforming to the curve, with both drawn within the same image. You would see gaps missing from the bar graph as it nears the edge of the curve. In the case of a lower quality recording with fewer samples, those gaps would be huge, and the results would be very apparent to anyone with eyes or ears. You can improve that by adding more samples (bars), but you will always have those gaps. At some point, you will reach an acceptable threshold.
Once that threshold has been achieved, you can cheat and add some additional filtering, which will give you either a better recording of the event, or it will “interpolate” the existing data to help refine the event into a better result.
At the end of the day, you determine the results by your own experience and judgement.