NURBS Curves & Surfaces

NURBS stands for Non-Uniform Rational B-Spline. Ignore the acronym for a second and picture the thing it describes: a smooth curve that follows, but usually does not touch, a set of control points. Drag a control point and the curve nearby shifts smoothly; the rest stays put. That local, predictable control is the whole appeal, and it is the reason NURBS became the standard for design - you can nudge a shape into exactly what you want without the whole curve lurching in response.

Four ingredients fully define the curve. The control points are the handles you move. The degree sets how smooth and how 'stiff' the curve is - degree 1 is straight segments (a polyline), degree 3 is the everyday smooth curve most modelling uses, and higher degrees are smoother but floppier and heavier. The knot vector is a non-decreasing list of numbers that parametrises the curve - it decides where each control point's influence begins and ends along the length; 'non-uniform' just means those spacings need not be equal, which is what lets NURBS bend where you want. The weights are the 'rational' part: give one control point a heavier weight and the curve is pulled closer to it. Uniform weights give an ordinary B-spline; unequal weights let NURBS describe conic sections - a true circle, ellipse or arc - exactly, which plain polynomial curves cannot.

There is one more useful idea hiding in the degree: a NURBS curve is built from many small polynomial pieces (spans) stitched together seamlessly, and the degree sets how many control points influence any given span - a degree-3 curve is locally shaped by four control points at a time. That is why editing one point only ripples out a short way, and why a Bezier curve (which you may meet elsewhere) is really just the simplest special case - a single-span curve where every control point affects the whole shape. NURBS generalise Beziers so you can chain long, editable curves without them going global and unruly.

You rarely edit knots or weights by hand. But knowing they exist explains a lot: why inserting a point changes the knot vector, why a 'rebuilt' curve looks subtly different, why a curve has a visible seam at its start point, and why degree matters when a loft or fillet comes out wrong.

A NURBS curve or surface is a continuous mathematical object. Ask for the point at parameter 0.5, or the tangent, or the exact area, and Rhino computes it from the equation - there is no resolution to run out of. Zoom in forever and a NURBS circle stays a perfect circle.

Contrast a mesh: a mesh approximates a shape with flat triangles or quads. It is light, fast and perfect for rendering, sculpting and 3D-printing, but a mesh circle is really a many-sided polygon, and its smoothness is baked in at creation. NURBS is the language of precise, manufacturable design - CAD, CNC, architectural surfaces - because it carries exact geometry and clean tangents all the way to fabrication. Meshes are the language of visualisation and organic form.

This is why Rhino is a NURBS modeller at heart (with strong mesh tools alongside): architectural and product geometry usually needs the exactness. It also explains a recurring workflow fact - you will often convert NURBS to mesh for analysis or rendering, but going the other way (mesh back to clean NURBS) is hard and lossy. Model in the exact world for as long as you can, and mesh only when you must.

The distinction matters more than it first appears when you reach Grasshopper. Some analysis and simulation tools - Kangaroo physics, many environmental workflows, game-engine exports - want meshes, because triangles are fast and predictable to compute on. Other operations - offsets, fillets, clean boolean solids, fabrication of a milled surface - want NURBS, because they need exact tangents and continuous geometry. A large part of being fluent later is knowing which representation each task expects and converting deliberately, at the right moment, rather than being surprised when a component refuses your geometry because it wanted the other kind.

A NURBS surface is the two-dimensional version of the same idea: a grid of control points in two directions, called u and v, spanning a smooth patch. Every single NURBS surface is topologically a rectangle - it has a u direction and a v direction and four sides - even when it looks like a doubly-curved blob. The faint lines you see flowing across a surface in Rhino are isocurves: curves of constant u or constant v. They are a display of the surface's internal grid, not real edges, and not something you can usually select and cut. New modellers constantly mistake isocurves for edges; they are not.

Real designs are rarely one clean rectangle, so surfaces get trimmed (a curve cuts a hole or a boundary while the underlying maths continues behind the trim) and joined. Join several surfaces along shared edges and you get a polysurface - Rhino's word for a shell of multiple surface faces stitched together. A closed, watertight polysurface with no gaps is a solid. Under the hood both are stored as a Brep (boundary representation): faces, edges and vertices describing a volume by its skin. Getting comfortable with this ladder - point, curve, surface, polysurface, solid - is most of what makes downstream modelling behave.

When two curves or surfaces meet, the quality of the join is described by geometric continuity. G0 (positional) means they simply touch - the endpoints coincide, but there can be a sharp kink, like two straight lines meeting at a corner. G1 (tangent) means they touch and share the same tangent direction at the seam - no kink, but the curvature can jump abruptly. G2 (curvature) means they touch, share tangent, and share the same curvature across the seam - the transition is truly smooth.

Why care? Because continuity is visible. Run a reflection or a zebra-stripe analysis across a G1 join and the stripes bend sharply at the seam; across a G2 join they flow through unbroken. Car bodies, product housings and premium architectural surfaces are held to G2 (or higher) precisely because the eye - and the highlight sliding across a real material - reveals anything less. For a rough massing model G0/G1 is fine; for a surface that will be milled and seen up close, aim for G2. Rhino's Match, BlendCrv and BlendSrf tools let you set exactly which level you want, and the analysis tools (Zebra, EMap, curvature graph) let you verify it honestly rather than trusting your eye.

Everything you will do in Grasshopper - dividing curves, lofting surfaces, panelising facades, evaluating points for an attractor - operates on this NURBS substrate. A Grasshopper Divide Curve component walks the curve by its parameter space, which is exactly the knot-driven parametrisation from the first section; Evaluate Surface samples a surface at a (u, v) coordinate; a loft's fairness depends on the degree and matching of its input curves. When a definition produces ugly, bunched or flipped geometry, the fix is usually upstream in the NURBS - a curve of the wrong degree, seams that do not align, surfaces whose u,v run opposite ways.

So this is not throat-clearing before the 'real' software. It is the grammar. Spend an afternoon dragging control points, changing degree, rebuilding curves and running Zebra on a few joins, and the abstract terms become physical intuitions. Every later module - data trees, geometry, facades, form-finding - assumes you can look at a curve and roughly know its degree, tell a surface from a polysurface, and predict whether two pieces meet cleanly. Get fluent here and the rest of the course stops fighting you.