Geometry
The Koch Snowflake: Infinite Perimeter, Finite Area
The Koch Snowflake is a fractal curve, introduced by Helge von Koch in 1904, built by an endlessly repeated rule: take an equilateral triangle and, on every edge, replace the middle third with an outward-pointing equilateral bump. Iterate forever. The astonishing payoff is a genuine paradox you can prove with a geometric series — the boundary has infinite length, yet it encloses only a finite area, exactly 8/5 of the original triangle. It is the cleanest example of a shape whose perimeter and area disagree about whether it is "big."- IntroducedHelge von Koch, 1904
- Construction ruleReplace middle third of each edge with an outward equilateral bump
- Perimeter growth×4/3 per step → ∞
- Limiting area8/5 · A₀ = 1.6 · A₀
- Fractal dimensionlog 4 / log 3 ≈ 1.2619
- RegularityContinuous everywhere, differentiable nowhere
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
The construction rule, stated precisely
Start with an equilateral triangle of side s₀; call its curve K₀. To get Kₙ₊₁ from Kₙ, apply the Koch rule to every straight segment: divide it into three equal parts, erase the middle third, and glue in the two other sides of an equilateral triangle that points outward. One straight segment of length ℓ becomes four segments each of length ℓ/3 — a jagged path of total length 4ℓ/3.
The snowflake curve K is the limit K = lim Kₙ as n → ∞. This limit genuinely exists: each point of Kₙ moves by at most a bounded, geometrically-shrinking amount at the next step, so the sequence of curves converges uniformly (in the Hausdorff sense) to a well-defined continuous curve.
Two bookkeeping facts drive everything. At step n the curve has Nₙ = 3·4ⁿ segments (we start with 3 edges, and each segment splits into 4), and each segment has length ℓₙ = s₀·(1/3)ⁿ. That single pair of formulas — the number of pieces multiplied by 4, their length divided by 3 — is the entire engine of the paradox.
Why the perimeter is infinite
The perimeter is just count × length: Pₙ = Nₙ · ℓₙ = 3·4ⁿ · s₀·(1/3)ⁿ = 3·s₀·(4/3)ⁿ.
Each step multiplies the perimeter by exactly 4/3. Since 4/3 > 1, the geometric factor (4/3)ⁿ grows without bound, so Pₙ → ∞. Concretely, with s₀ = 1 the perimeters run 3, 4, 16/3 ≈ 5.33, 64/9 ≈ 7.11, 256/27 ≈ 9.48, 1024/81 ≈ 12.64, … — never settling, always ×4/3.
Intuition. Every time you zoom in, the boundary is not getting smoother — it is getting crinklier, and it does so at the same relative rate at every scale. Smoothing would let length converge; this self-similar roughening feeds in a fixed 33% surplus of length at every step forever. It is the same reason a real coastline gets longer the finer your measuring stick: there is detail at every scale. The Koch curve is the idealized coastline where that never stops.
Why the area is finite — worked out
Let A₀ be the area of the starting triangle (A₀ = √3/4 · s₀²). At step 1 we bolt one small triangle onto each of the 3 edges; at step 2, one onto each of the now 3·4 = 12 edges; in general at step n we add one bump per edge of Kₙ₋₁, i.e. 3·4ⁿ⁻¹ new triangles.
Each new triangle at step n sits on a segment of length s₀·(1/3)ⁿ, so its area is A₀·(1/9)ⁿ (area scales as the square of length, and 1/3² = 1/9). Total added area at step n:
ΔAₙ = (3·4ⁿ⁻¹) · A₀·(1/9)ⁿ = A₀ · (3/9) · (4/9)ⁿ⁻¹ = A₀ · (1/3) · (4/9)ⁿ⁻¹.
Now sum the whole thing:
- A = A₀ + ∑ₙ₌₁^∞ ΔAₙ = A₀ + A₀·(1/3)·∑ₖ₌₀^∞ (4/9)ᵏ.
- The bracketed geometric series has ratio 4/9 < 1, so it converges to 1/(1 − 4/9) = 9/5.
- A = A₀·[1 + (1/3)·(9/5)] = A₀·[1 + 3/5] = 8/5 · A₀.
Numeric check. Running partial areas (A₀ = 1): 1 → 4/3 ≈ 1.333 → 40/27 ≈ 1.481 → 376/243 ≈ 1.547 → 1.577 → 1.590 → 1.595 …, tightening toward 1.600 = 8/5. The perimeter runs off to ∞ while the area is trapped below 1.6·A₀.
The reconciliation: length and area are different questions
The paradox feels illegal until you notice that perimeter and area are answers to different questions. Length is a one-dimensional measurement; area is two-dimensional. The Koch boundary is too rough to be one-dimensional but too thin to be two-dimensional — it lives in between.
That "in between" is made precise by the fractal (Hausdorff) dimension. A self-similar set made of N copies of itself each scaled by r has dimension D = log N / log(1/r). The Koch curve is 4 copies at scale 1/3, so D = log 4 / log 3 ≈ 1.2619. Because 1 < D < 2, the correct 1-D measure (length) is +∞ and the correct 2-D measure (area) of the curve itself is 0 — the finite 8/5·A₀ is the area of the region enclosed, not of the boundary.
The mental picture: the boundary is squeezed into an ever-thinner tube. Its length blows up, but the tube never expands outward past a fixed frame — every bump added at step n lies inside the previous bump's triangle-shaped neighborhood, and those neighborhoods have areas that shrink like (4/9)ⁿ. Bounded total area, unbounded total length: no contradiction, just two coordinates of "size" pointing different ways.
Where it matters, a generalization, and a pitfall
Why it matters. Von Koch built this in 1904 as a hand-drawable answer to a shocking result of Weierstrass: a curve can be continuous everywhere yet have a tangent nowhere. The Koch curve is exactly that — you can trace it, but at no point does it have a well-defined slope (each point is a corner at infinitely many scales). It became a founding example of fractal geometry and the standard toy model for scale-invariant roughness — coastlines, mountain silhouettes, cloud edges, and antenna design (Koch-shaped antennas pack more electrical length into a small footprint).
Generalization. Replace the middle-third bump with a middle-δ segment raised to angle θ, or use a different generator polyline, and you get the whole family of de Rham / Koch-type curves. A generator that makes N copies at scale r has dimension log N / log(1/r); pushing the bump higher raises the dimension toward 2, giving space-filling-like curves.
Pitfall. A frequent error is to "prove" a finite perimeter by summing side lengths as if they were areas — but Pₙ = 3(4/3)ⁿ has ratio 4/3 > 1 and diverges, while the area series has ratio 4/9 < 1 and converges. Same shape, two series, opposite fates: always check the ratio. A second misconception is thinking the finite "8/5" is the area of the curve; the curve (a 1.26-dimensional set) has zero 2-D area — 8/5·A₀ is the region it bounds.
| Property | Koch snowflake | Cantor set | Sierpiński triangle |
|---|---|---|---|
| Build rule | Add middle-third bump | Delete middle third | Delete middle sub-triangle |
| Copies N / scale r | N = 4, r = 1/3 | N = 2, r = 1/3 | N = 3, r = 1/2 |
| Dimension log N / log(1/r) | log4/log3 ≈ 1.262 | log2/log3 ≈ 0.631 | log3/log2 ≈ 1.585 |
| Length / measure | Infinite perimeter | Zero length | Zero area |
| Encloses area? | Yes, finite (8/5·A₀) | No (subset of a line) | Area → 0 |
Frequently asked questions
How can the perimeter be infinite if the snowflake fits inside a small circle?
Fitting inside a bounded region caps the enclosed area, not the boundary length. A curve of finite diameter can still have infinite length if it wiggles at every scale. Pₙ = 3·(4/3)ⁿ grows by 4/3 each step and diverges, even though every point stays within a fixed disc around the original triangle.
What is the exact limiting area?
Exactly 8/5 of the starting triangle's area, i.e. 1.6·A₀. Summing the added bumps gives A₀·[1 + (1/3)·∑(4/9)ᵏ] = A₀·[1 + (1/3)·(9/5)] = 8/5·A₀. With A₀ = √3/4·s₀², the snowflake area is (2√3/5)·s₀².
What is the fractal dimension of the Koch curve?
D = log 4 / log 3 ≈ 1.2619. The curve is self-similar as 4 copies each scaled by 1/3, and the similarity dimension is log(number of copies)/log(inverse scale). Being strictly between 1 and 2 is exactly why its length is infinite and its area is zero.
Is the Koch curve differentiable anywhere?
No. It is continuous everywhere but differentiable nowhere. At every point and every scale there is another corner, so no tangent line exists at any point — it is a concrete example of the continuous-but-nowhere-differentiable curves Weierstrass had constructed analytically in the 1870s.
Why divide by 9 for area but by 3 for length at each step?
Lengths scale linearly, so a bump segment is 1/3 as long. Areas scale as the square of length, so each new triangle is (1/3)² = 1/9 the area of one at the previous scale. That squaring is precisely what turns the diverging length series (ratio 4/3) into a convergent area series (ratio 4/9).
Who discovered it and when?
Helge von Koch, a Swedish mathematician, in a 1904 paper. He wanted a simple, explicitly geometric example of a continuous curve with no tangent — an elementary alternative to Weierstrass's analytic function. The snowflake (three Koch curves on a triangle) followed as the natural closed version.