Rebuilding Pythagoras from scratch (no memorising). Put four copies of our triangle around a tilted square whose side is the chord Li. The big outer square has side Δxi+Δyi. Its area equals the tilted inner square (Li2) plus the four triangles (4×21ΔxiΔyi):
big square(Δxi+Δyi)2=tilted squareLi2+four triangles4⋅21ΔxiΔyi
Expand the left side: (Δxi)2+2ΔxiΔyi+(Δyi)2. The 2ΔxiΔyi cancels the triangles, leaving
Li=(Δxi)2+(Δyi)2.
(Δxi)2 — the sideways travel, squared.
(Δyi)2 — the vertical travel, squared.
— undoes the squaring to return an actual length. See Pythagorean theorem.
Li=(Δxi)2+(Δyi)2=(Δxi)2(1+(Δxi)2(Δyi)2)=1+(tilt of the chordΔxiΔyi)2Δxi
We could pull Δxi out of the root because Δxi>0 (we always walk left-to-right — this is exactly the "no doubling back, advance rightward" assumption from Step 1 — so (Δxi)2=Δxi, no minus sign).
The 1 is the leftover from the sideways leg — it is what stops us forgetting the horizontal travel.
ΔxiΔyi = vertical-move ÷ right-move = the slope of the chord (a "secant slope"). If the chord falls, this ratio is negative — but it gets squared next, so again the sign washes out.
Flat curve (f′=0). Vertical leg Δyi=0; the triangle collapses to a horizontal segment. Integrand =1+0=1, so L=∫ab1dx=b−a — exactly the width. ✓ A flat line's length is its horizontal span.
Vertical wall (f′→∞). The slope blows up; the y=f(x) formula chokes because a vertical piece has Δxi=0 (nothing to pull out). Fix: flip to the x=g(y) form and integrate in y instead — the vertical leg becomes the "safe" direction.
Corner (slope jumps). At a sharp kink f′ doesn't exist — this violates MVT's differentiability condition at that one point. Since one bad point has zero width, the integral is unbothered — just split the interval at the corner and add the two pieces.
Read the film strip left-to-right: curve → chords → one triangle → pull out dx → tangent matches chord → strips sum to an integral. Every arrow is one step above.
Recall Feynman: the whole walkthrough in plain words
You want the length of a bumpy path. You can't measure a curve directly, so you lay tiny straight sticks end-to-end along it (Step 2). Each stick is the long side of a little right triangle whose legs are "a bit rightward" and "a bit up or down"; the corner-square trick (Pythagoras) says the stick's length is right2+vertical2 — and because the vertical bit gets squared, it doesn't matter if the curve climbs or falls (Step 3). To collect everything into a sum over rightward steps, you factor the rightward bit out and are left with 1+slope2 times the rightward step (Step 4). "Slope of the stick" isn't quite the curve's real slope, but the Mean Value Theorem — as long as the curve is smooth (continuous and differentiable) — guarantees the real tangent somewhere inside is exactly as tilted, so you replace it by f′ (Step 5). Now every stick reads 1+(f′)2Δx; adding infinitely many infinitely short ones is what the integral sign means, giving L=∫ab1+(f′)2dx (Step 6). Test it on a flat road (length = width), a vertical cliff (integrate in y instead), and a sharp corner (split there) — it survives all three (Step 7). Same ds=dx2+dy2 later builds surfaces of revolution, parametric and polar lengths.