Curves & Surfaces in Grasshopper

There are two everyday ways to build a curve from points, and the difference matters. Interpolate Curve (the Interpolate / IntCrv component) draws a smooth curve that passes through every point you give it - use it when the points are positions the curve must hit (a path traced on site, a section you measured). Nurbs Curve instead treats your points as control points: the curve is pulled toward them but generally does not touch them, like a magnetic cage shaping a smooth line. Control-point curves give you sleeker, more controllable fairness; interpolated curves give you precision through known points. Knowing which you want saves hours of confusion about 'why isn't my curve touching the points'.

Under both sits NURBS - Non-Uniform Rational B-Splines - the smooth, resolution-independent curve maths Rhino is built on. You don't need the equations, but three ideas help: degree (roughly, how smooth and how much each control point's influence spreads - degree 3 is the sane default), control points (the handles that shape it), and knots/domain (the internal parameterization, which we're about to tame). The payoff of NURBS is that a curve stays perfectly smooth and exact at any zoom, and can be evaluated analytically - which is exactly what the next components exploit. There's a fuller treatment in the Rhino & NURBS module; here we just use it, trusting that a degree-3 curve through or near a handful of points will be smooth, fair and ready to evaluate.

A curve you can't query is half useless. Evaluate Curve takes a curve and a parameter t and returns the point at that spot plus the tangent vector there - 'give me the position and heading 50% along'. Divide Curve chops a curve into a chosen number of equal-length segments and returns the division points, the tangents, and the parameters - this single component is behind an enormous amount of parametric work: fence posts, facade mullions, sampling a path for an array. Divide Distance and Divide Length are cousins for when you want a fixed spacing rather than a fixed count.

Now the idea that trips up everyone: a raw curve's parameter domain is not 0-to-1. It runs over some arbitrary interval baked in when the curve was made (say 0 to 43.7). So asking for t = 0.5 does not give you the middle - it gives you a spot right near the start. The fix is Reparameterize: right-click a curve input and choose 'Reparameterize', or use the component, and the domain is remapped to a clean 0 to 1. Now 0 is the start, 1 is the end, 0.5 is the middle, and any 0-to-1 value (from a slider, an attractor, a graph mapper) addresses the curve intuitively. Reparameterizing is a tiny habit that removes a whole category of bugs - build it in early.

Surfaces in Grasshopper are usually generated from curves, and four components cover most of it. Extrude pushes a curve (or surface) along a vector - a wall from a plan line, a mullion from a rectangle. Loft stretches a smooth surface across a list of section curves in order - the workhorse for hulls, roofs, and any form defined by a series of profiles; feed it three-plus sections and it skins them (watch the seam directions - flipped or mis-ordered sections give you a bow-tie, a classic first-loft mistake). Sweep1 runs a profile curve along a single rail; Sweep2 runs it between two rails, letting the profile scale to fit - perfect for a handrail, a cornice, a tapering canopy edge. Revolve spins a profile around an axis for anything rotationally symmetric.

The mental model is the same as with curves: these are construction components driven by parameters. Because the inputs are curves you built parametrically, the surface updates when they do - move a control point on a section curve and the loft re-skins live. And every surface these produce is, again, NURBS: smooth, exact, and - importantly for the next section - carrying its own internal 2D coordinate system you can sample.

Here is the concept that separates people who can panelize a facade from people who fight one. Every NURBS surface has a built-in 2D coordinate system stretched across it, called UV space. Think of the surface as a rubber sheet with a printed grid: u runs one way, v runs the other, and any point on the surface has a (u, v) address - regardless of how the surface bends in 3D. It's latitude and longitude for that specific surface.

This is what lets you sample a surface systematically. Evaluate Surface takes a (u, v) and returns the 3D point there, the normal vector (which way is 'out'), and the local frame - the plane you'd place a panel on. Divide Surface hands you a whole UV grid of points and normals in one shot: a 20-by-12 lattice of panel anchors, each with its outward direction, ready to array glazing or shading. Reparameterize the surface so UV runs 0-to-1 (same habit as curves) and a slider from 0 to 1 walks smoothly across it. Almost every parametric facade in Module 6 is, at heart, 'Divide Surface into a UV grid, then place an oriented panel on each cell's frame'. Understand UV now and that module is mostly bookkeeping. One honest caveat: UV spacing is even in parameter space, not always in real distance - on a stretched or trimmed surface the grid can look uneven, which is a known quirk you handle with equalized division or by rebuilding the surface.

A parametric surface is only as trustworthy as the curves and habits behind it, so a few disciplines separate definitions that survive editing from ones that shatter. First, treat your generating curves as the design and keep them clean: consistent direction, sensible degree, no accidental kinks or duplicate control points. A messy section curve makes a messy loft, and no amount of downstream fiddling recovers it - fix the input, not the output. Second, reparameterize early and everywhere - curves and surfaces - so every domain speaks the same 0-to-1 language; mixing native and reparameterized domains in one definition is a reliable way to get points that land almost-but-not-quite where you expect.

Third, mind the seam and the UV directions. Every closed surface has a seam (where it wraps back on itself), and every surface has a U and a V that may not be the way you assume - a facade you expected to divide '20 wide by 8 tall' can come out '8 wide by 20 tall' because U and V are swapped. Display the surface's isocurves and a small UV indicator while you build so you see which way is which, and use Flip or Swap UV when needed rather than guessing. Fourth, remember that surfaces you sample feed straight into the next stages of the course: the points from Divide Surface become attractor targets in Module 5, panel anchors in Module 6, and mesh vertices when you hand off to analysis or fabrication in Module 9. Because of that, a well-parameterized, evenly divided surface pays dividends far downstream, while a sloppy one quietly corrupts everything built on it. Getting curves and surfaces right here is not busywork - it is the foundation the expressive modules stand on.