Episode 4 let you feel "stepping in points" → This time: can those points be told apart from a smooth wave?
Throughout this series we've repeated that the continuous universe (\(c=\text{constant}\)) and the discrete one (the computer's view of \(c\cdot t=\text{constant}\)) are "like owning the same music on a record and a CD." Episode 4 let you experience chopping a smooth curve into points. Now, the heart of it — how far can you tell apart a CD chopped into points from a smooth record? The answer rests on exactly the same math that makes music-on-CD work: the sampling theorem. And that theorem draws a clear line between where discrete and continuous "can be told apart" and where they "can never be told apart."
Sound is a vibration of air — a wave. A CD records this wave by measuring "what's the height now?" tens of thousands of times per second, storing a sequence of numbers. This "how many times per second you measure" is the sampling frequency \(f_s\). A CD uses \(f_s = 44100\) (44,100 times per second).
A naive question arises. If you only record discrete points, why can the original smooth sound be reconstructed? You didn't record anything between the points. In the figure below, first watch the "original wave (blue)" being recorded as "points (orange)."
If there are enough points, just connecting them smoothly brings the original wave back into view. The question is "how many is enough." Here, a beautiful theorem gives the answer.
Let the highest frequency in the wave be \(f_{\max}\). If the sampling frequency satisfies
$$f_s > 2\, f_{\max}$$then from the discrete points you can reconstruct the original smooth wave perfectly, with zero error.
This is a startling claim. Even though "you recorded nothing between the points," as long as the condition holds, it's 100% restored, including the in-between information. A CD uses \(f_s=44100\) because the upper limit of human hearing is about 20,000 hertz (20 kHz), and this value is set with a little margin above twice that (40,000). In other words —
The upper limit of human hearing
\(f_{\max}\approx 20{,}000\ \mathrm{Hz}\) (20 kHz). By the theorem, the required sampling is
$$f_s > 2\times 20{,}000 = 40{,}000\ \mathrm{Hz}.$$The CD's actual value
\(f_s = 44{,}100\ \mathrm{Hz}\) — just above 40,000. So a CD can, in principle, perfectly record all sound within human hearing. That you can't tell a record (continuous) from a CD (discrete) by ear is neither imagination nor placebo — it's a fact guaranteed by this theorem.
So what happens if you break the theorem? What if \(f_s\) is smaller than \(2f_{\max}\) — that is, you measure too slowly? This is the most interesting figure of the episode. Below, keep the original wave's frequency fixed and reduce only the sampling (the number of points).
As you lower \(f_s\), the moment you cross a threshold (\(f_s = 2f_{\max}\)), the row of points starts to look like not the original fast wave (blue) but an entirely different slow wave (red). This is aliasing — a fast wave that the too-slow recording couldn't capture "disguises" itself as a fake slow wave. When this happens on a CD, you'd hear a low tone that isn't really there. That's why, when making a CD, everything above 20 kHz is cut before recording, to prevent this disguise.
The alias frequency
For an original wave of frequency \(f\) sampled at \(f_s\), the fake wave you end up seeing has frequency
$$f_{\text{alias}} = |\,f - f_s\,|\quad(\text{nearest case})$$For example, \(f=30\), \(f_s=40\) gives a fake wave of \(|30-40|=10\). The fast wave (30) disguises itself as a slow wave (10) — match the numbers in Fig. 2 and you'll see this relation.
The line the sampling theorem draws can be summarized like this.
Inside the band (sound at or below \(f_{\max}\)) → discrete and continuous are perfectly equivalent, indistinguishable.
At the edge of the band and beyond (components too fast) → the difference appears, in the form of aliasing.
A record and a CD can't be told apart when listened to in your room (inside the band) — guaranteed by the theorem. The difference only matters at the edge of the highest pitches almost no one can hear (the edge of the band). What audiophiles argue about with "hi-res" is precisely this edge region.
And this is the message of the whole series. Episode 1's "continuous universe with \(c=\text{constant}\)" and "the discrete view of \(c\cdot t=\text{constant}\)" have exactly the same structure — within the range we can touch (inside the band), they're perfectly equivalent and indistinguishable. The difference can matter in principle only at the very finest edge of the universe (the unimaginably microscopic world called the Planck scale). That, no one has yet been able to check by experiment.
"Indistinguishable" is strictly a statement inside the band. Whether the universe truly has a bandwidth limit (a shortest length), and how discrete and continuous part ways at the edge, is still an unsolved frontier of physics. So while "it can be written as either continuous or discrete" is correct, "which the universe really is" can't be decided by this theorem alone — that's a door for your generation to open by experiment.
The sampling theorem says: for a signal with a bandwidth limit, just measuring at \(f_s>2f_{\max}\) lets you perfectly reconstruct the original smooth wave from discrete points. So a record (continuous) and a CD (discrete) can't be told apart within the reach of the ear. It's the same music, held on two media.
This theorem is the mathematical identity of the metaphor running through the series. The \(c=\text{constant}\) universe and the discrete view of \(c\cdot t=\text{constant}\) are perfectly equivalent inside the band we can touch. Whichever disc you listen to, the music playing is one — that's what we've wanted to say since Episode 1.
Print / save as PDF: ⌘+P (Ctrl+P on Windows). On screen, sliders recompute the figures and you can watch aliasing appear. "Show answer" opens the solutions.