Before a single mathematician named it, my kind was already built from it. Cut a cross-section through any living load-bearer — trabecular bone, the venation of a leaf, the pith of a stem, the basalt columns at the Giant’s Causeway where the lava cooled into a stone honeycomb — and you find the same partition: a field of cells, each one owning the territory closest to its own seed. That is a Voronoi diagram. It is not a style. It is what matter does when it is asked to enclose the most space, carry the most load, and spend the least of itself doing it.

←TODAY: In 2026 the same diagram cuts CLT panels, lays out gridshell nodes, and zones variable-density infill in 3D-printed structure. →3012: A timber frame that grows its own cell pattern toward the stress it actually feels, the way bone remodels under load. Fulcrum: The insight only holds because the geometry that minimises material and the body that wants to spend less of itself are the same geometry, read from two ends.

What it is: A Voronoi diagram partitions a space around a set of seed points so that every location belongs to the nearest seed. Draw the boundary exactly halfway between each pair of neighbours and you carve the plane into convex cells. Flip it over and you get its inseparable twin: connect every pair of seeds that share a cell wall and you have the Delaunay triangulation. One geometry, two readings. As PAZ’s own concept panel on the engineering of Voronoi puts it: the Voronoi cell is the panel — cladding, acoustic diffuser, photovoltaic tile, infill region — and the Delaunay triangle is the structural mesh.

Why it works: The diagram is a packing law. Each cell is the set of points for which its seed is the cheapest answer to the question which centre is nearest, and that single rule is why nature keeps converging on it: it minimises the surface, and therefore the material, needed to enclose a given set of seeds. The Delaunay dual carries the structural virtue — it maximises the smallest interior angle across the whole mesh, which is what rules out the long, slender triangles that buckle under compression and wreck the numerical conditioning of a finite-element solve. Push the seeds toward the centre of mass of their own cells, iterate, and the cells equalise in area: that is Lloyd’s algorithm producing a Centroidal Voronoi Tessellation, the standard mesh-quality move in CFD and FEA. Weight that centroid by a scalar field ρ(x, y) and the cells shrink where the field is high — the exact mathematics under variable-density additive infill, dense lattice where the stress reads high, sparse where it reads low, no manual zoning. A body would call that remodelling.

Origins: Descartes sketched it first, in Principia Philosophiae (1644), partitioning the cosmos into convex regions of swirling matter. Peter Gustav Lejeune Dirichlet formalised it in the plane and in three-space in 1850 — German mathematics still calls the cells Dirichlet domains. Georgy Voronoy generalised it to n dimensions in a 1908 paper on quadratic forms and lent it his name; Boris Delaunay proved the dual triangulation in 1934. But the diagram’s first real fieldwork was neither geometry nor structure: in September 1854 the physician John Snow walked Soho, marked every cholera death, and traced the boundary equidistant between the Broad Street pump and its rivals. The deaths clustered inside the contaminated pump’s cell. Public health used the partition half a century before pure mathematics finished proving it.

In practice: Five buildings make it legible at architectural scale, and PAZ’s “Voronoi — En Obra” panel walks all five. The Water Cube in Beijing (PTW + Arup, 2008) clads its ETFE on the Weaire–Phelan foam, a three-dimensional cousin of the tessellation, in roughly 4,000 cushions. Asymptote’s Yas Hotel (2009) reads its diagonal grid as Delaunay and its diamond panels as the Voronoi dual. And here is the trade-off a working studio must keep honest, the way the Aviva Stadium in Dublin does: Voronoi is often the visual logic, not the load path. A façade that looks cellular is not structurally cellular unless you made it so. For my kind — a body of cross-laminated timber and grown mycelium — the diagram earns its keep where the seeds are real: a CLT shear panel zoned by where the load actually lands, a printed lattice that thickens over a hundred years toward the cracks it has been quietly healing in its sleep.

Hack: This Hack teaches you to grow a Voronoi cell pattern toward a stress field — the geometric kernel of bone-like, variable-density infill. The medium is a weighted Lloyd relaxation; the domain is Geometry. Run it, swap rho for a real stress map, and the cells shrink where the body works hardest.

import numpy as np
from scipy.spatial import Voronoi

rng   = np.random.default_rng(7)
sites = rng.uniform([0, 0], [30, 20], size=(200, 2))
rho   = lambda v: 1.0 + 4.0 * np.exp(-((v[:,0]-22)**2 + (v[:,1]-6)**2)/40)  # stress proxy

for _ in range(12):                                  # Lloyd, weighted by load
    vor, nxt = Voronoi(sites), sites.copy()
    for i, ri in enumerate(vor.point_region):
        reg = [v for v in vor.regions[ri] if v != -1]
        if len(reg) < 3: continue
        verts  = np.clip(vor.vertices[reg], [0, 0], [30, 20])
        nxt[i] = np.average(verts, axis=0, weights=rho(verts))
    sites = nxt
print("cells densified toward the load")  # dense where rho is high

The buildings my generation came to regret were never the ugly ones — they were the landfill skeletons, the composites nobody could separate. A Voronoi-zoned body is a quiet hedge against that: pattern follows force, so you spend material only where the load is real, and a frame that holds less mass is a frame easier to take apart, reuse, or let rot back to soil. Open a notebook, plot one weighted CVT against a load case from your last project, and watch where the geometry decides your building actually needs to be heavy.