Derivatives & Integrals as Filters
Here's a second way to see differentiation and integration — not as slope and area, but as filters that clean up (or mess up) a signal. This is how engineers actually think about them.
The logic in one breath:
• Differentiation subtracts neighboring values. Things that barely change between neighbors cancel out; only the fast wiggles survive. So differentiation keeps fast changes and throws away slow ones — a high-pass filter.
• Integration adds neighboring values (like averaging them). Fast up-down jitter cancels in the averaging; the slow trend survives. So integration smooths — a low-pass filter.
Real-world picture: the shock absorbers on a bike are integrators — they smooth out the fast bumps in the road (low-pass) so you feel only the slow hills. Differentiation is the opposite: it would amplify every pebble.
See it for yourself. Below is a smooth wave with noisy jitter added. Differentiating it makes the noise explode; integrating (averaging) it washes the noise away. Crank the noise slider and roll fresh noise with the 🎲 button.
Same signal, two operations. Differentiating (subtracting neighbors) keeps only what changes fast — so it amplifies the jittery noise: a high-pass filter. Integrating (adding neighbors) keeps only what changes slowly — it washes the noise out and leaves the smooth wave: a low-pass filter. Crank the noise up and watch the red plot go wild while the green stays calm.
Why this matters in the lab: real neural recordings are noisy. If you want the slow trend, you smooth (integrate-like). If you want to catch sharp spike onsets, you differentiate — but you'll amplify noise too, so you trade one problem for another. These add-and-subtract operations are literally the basis of every digital filter.