K-means with k = 3 alternates two steps: each point joins its nearest centroid, then each centroid moves to the mean of its points.
Neither step can raise WCSS = ∑ |xᵢ − c(xᵢ)|², since the mean minimises summed squared distance. The sum only falls, and with finitely many partitions the loop must stop: here 326.1 falls to 34.8 in 7 steps.
The stop is only a local minimum, so k-means is usually run from several starts.
A streamline of a vector field F is the path of a particle whose velocity is always the arrow beneath it: dx/dt = F(x, y).
When F = (∂H/∂y, −∂H/∂x), every streamline lies on a level curve of H: x³ + y³ for (y², −x²), x² + y² for (−y, x), cos x − cos y for (sin y, sin x). The field (x − y, x + y) is z ↦ (1 + i)z, so z₀e⁽¹⁺ⁱ⁾ᵗ spirals out.
The first three have zero divergence and carry every area unchanged; the spiral stretches areas by e²ᵗ.