Diagrids & Structural Patterns

Start with the humblest fact in structures: a triangle is rigid, a rectangle is not. Pin four bars into a square and it racks - it collapses sideways into a parallelogram - because the joints can rotate. Pin three bars into a triangle and it cannot change shape without changing a bar's length, which the bars resist directly in tension or compression. This is why every stiff lattice, from a bridge truss to a bicycle frame, is built from triangles.

A diagrid (diagonal grid) exploits this by wrapping a building in triangulated members instead of the usual verticals and horizontals. The diagonals carry both the gravity load and the lateral load (wind, seismic) as axial forces in the members, so the structure barely needs separate bracing or perimeter columns. The famous result is dramatic material savings: diagrid towers such as London's 30 St Mary Axe (the 'Gherkin') and the Hearst Tower in New York are often cited as using markedly less structural steel than an equivalent conventional frame, precisely because triangulation turns bending problems into axial ones - the same efficiency logic as the funicular arch, now in a lattice.

A space frame extends the idea into three dimensions: a thick, two-layer triangulated lattice (think of a big exhibition-hall roof) where every member is short and axially loaded, spanning huge distances with little depth. Whether it is a flat space frame, a curved diagrid shell, or a triangulated facade, the underlying move is identical - make the load path a network of triangles so members push and pull rather than bend.

It is worth internalising why this is such a big deal. Bending wastes material because only the extreme fibres of a section work hard while the core barely earns its keep; axial force uses the whole cross-section evenly. Turning a structure from a bending problem into an axial one is therefore one of the highest-leverage moves in all of engineering, and triangulation is the geometric trick that does it. The diagrid simply applies that trick to a whole building envelope.

Here is where computational design earns its keep. To a Grasshopper definition, a diagrid is simply a pattern generated on a surface. The recipe is generic: take your (often form-found) surface, evaluate a grid of UV points across it with Divide Surface or Isotrim/Surface Box, then connect those points diagonally instead of orthogonally to make triangles. Plug-ins like LunchBox package this directly - its 'Diagrid Structure' and various panel components take a surface and a U/V count and hand back the diagonal member lines and panels ready to build on.

Because it is parametric, every property of the diagrid is a dial. Density is the U and V counts - more subdivisions give more, smaller triangles. Angle is how the diagonals are drawn - a steeper diagonal changes how much each member leans toward carrying vertical versus lateral load. You can even let the density vary across the surface using the attractor and field logic from Module 5: tighten the grid where an engineer tells you forces concentrate (around a base, an opening, a corner) and open it where they do not. That is structure and pattern responding to the same parameter - the essence of expressing structure as architecture.

Crucially, because the members come out as clean lines and the nodes as points, the same definition that designs the diagrid also produces the geometry to analyse and fabricate it: line members feed straight into Karamba3D for structural checking, and node points feed into the connection and fabrication drawings. One parametric pattern, driving design, analysis and fabrication together.

The single most important design conversation in a diagrid is the trade between grid density and member size, and it is genuinely a trade - there is no free lunch. Make the grid denser (more, smaller triangles) and each member carries a smaller share of the total load, so members can be slimmer and the facade reads as fine and delicate. But you now have far more nodes, and nodes - the connections where members meet - are the expensive, labour-intensive, structurally critical part of any diagrid. Make the grid coarser (fewer, bigger triangles) and you have fewer nodes to fabricate, but each member and each joint carries more load, so members get chunky and the connections become heavy castings.

So the density parameter is really choosing where the cost and the expression go: many slender members and many nodes, or few massive members and few nodes. Add to this the glazing or cladding panels that sit in the triangles - denser grids mean more, smaller (and often cheaper, flatter) panels but more framing; coarser grids mean large panels that may need to be curved or specially made. And node buildability governs everything: a node where six members meet at awkward angles is a bespoke fabrication problem, so designers often constrain the geometry to keep node types few and repeatable.

This is exactly the kind of multi-way trade computational design is built to explore - and, in Module 8, to optimise. You can sweep density and read off member count, node count and panel count for each option. But note the sharp line: the definition tells you how many members and nodes and how long each is; it does not tell you how thick they must be. Member sizing - cross-sections, wall thickness, connection capacity, buckling - is structural engineering, done with a tool like Karamba3D and, for anything real, a qualified engineer. Parametrics lays out the pattern; the engineer sizes the steel.

The reason diagrids are everywhere in contemporary architecture is that they let the structure be the architecture. There is no separate cladding hiding a frame; the triangulated members are what you see, and their rhythm, density and angle are the building's visual language. This is an old and honest idea - Gothic vaults, Buckminster Fuller's geodesic domes, Nervi's ribbed concrete - given a new, tunable life by parametric tools. When you vary a diagrid's density with a field, you are composing a facade and routing structural forces with the same gesture, which is a genuinely satisfying place for design and engineering to meet.

But 'expressing structure' carries a responsibility to be truthful about it. A triangulated pattern sprayed onto a surface purely for looks - members that carry nothing, nodes that connect decoratively - is structural decoration, not structure, and it is worth being clear with yourself and your client about which you are doing. Real structural diagrids earn their expression by actually carrying load; the pattern and the force path are the same thing. That honesty is also what keeps the collaboration with the engineer productive: if the members you are drawing are meant to work, their geometry has real consequences, and the engineer's feedback (this angle is too shallow, this node can't be built, this density won't brace the corner) should reshape your parameters.

So treat a diagrid as a hypothesis about how the building stands, expressed as a pattern you can tune - and treat the engineer as the one who confirms and sizes it. The most beautiful diagrids are not the ones with the cleverest pattern; they are the ones where the pattern and the structure are the same true thing, developed together. Computational design gives you the language to have that conversation precisely.