Geometric Modeling

Geometric modeling is the part of computer graphics that invents ways to describe shape inside a computer so that the shape can be drawn, analyzed, manufactured, or simulated.

10 frameworks, from 1960 to today · Computer Graphics · Computer Science

still a working view ended · each leads to its entry below

Overview

Geometric modeling is the part of computer graphics that invents ways to describe shape inside a computer so that the shape can be drawn, analyzed, manufactured, or simulated12. Its central problem is representation: a physical object—a car body, a teapot, a bone, a gearbox—has to be turned into numbers and equations that a machine can store and manipulate, and different representations make different things easy. A representation that makes it easy to draw a smooth fender may make it hard to compute whether a drill will pierce the metal; a representation that makes it easy to test solidity may make it awkward to sculpt a free-form curve. Much of the field's history is the story of building representations that trade these conveniences off, and then of building bridges between them. What is at stake is not just pictures but the objects themselves: the same models feed animation, engineering analysis, 3D printing, and medical planning, so a modeling choice made in a design studio can propagate all the way to a factory floor.

Curves and patches for car bodies and aircraft

The earliest practical need came from manufacturing. In the 1950s and 1960s, aircraft and automobile companies wanted to describe smooth free-form surfaces—surfaces with no simple algebraic formula, shaped by aesthetic and aerodynamic judgment—in a form a computer could store and a milling machine could cut. The French engineer Pierre Bézier, working at Renault, developed the curve representation now called the Bézier curve, in which a curve is defined by a small set of control points that pull it into shape; the curve passes through the first and last points and is drawn toward the others without necessarily touching them. Bézier published his work in the 1960s, and the same idea was developed independently around the same period by Paul de Casteljau at Citroën, whose algorithm for evaluating such curves was kept as a company secret for years. The appeal was direct manipulation: a designer could move a control point and watch the curve respond predictably, which matched the way stylists already worked with physical splines.

The other foundational idea came from Steven Coons at MIT, who in the 1960s described how to build a curved surface patch from its four boundary curves—a Coons patch. This let designers stitch patches together along shared edges, so a complex surface could be assembled from manageable pieces. Bézier patches and Coons patches together established the patchwork approach: model a surface as a quilt of smooth pieces. The problem this left was continuity—how to make neighboring patches meet without a visible crease—and that question drove the next generation of work.

B-splines, NURBS, and the standardization of free-form shape

Bézier curves have a limitation: moving one control point changes the whole curve, and joining many curves smoothly takes care. The B-spline, developed in the mid-twentieth century and brought into computer-aided design in the 1970s, solved this by making each control point influence only a local stretch of the curve, and by building smoothness into the representation itself. A B-spline curve is defined by control points and a knot vector—a list of parameter values that says where each piece of the curve begins and ends—and the mathematics guarantees a chosen degree of continuity across the joins. This made it practical to model long, complicated curves and surfaces as single objects rather than hand-stitched patches.

The generalization that came to dominate industry is the NURBS curve and surface: a Non-Uniform Rational B-Spline, which adds weights to the control points so that the representation can describe exact circles, ellipses, and other conic sections—shapes that ordinary polynomial splines can only approximate. NURBS were formalized and popularized in the 1970s and 1980s, with Kenneth Versprille's doctoral work on rational B-splines among the key steps, and they became the standard exchange format for free-form geometry in CAD. The reason is telling: a single representation could hold both the free-form styling surfaces and the precise engineering features (a cylindrical hole, a conical chamfer) that a mechanical part needs. The IGES and later STEP data-exchange standards carried NURBS geometry between incompatible software systems, and NURBS remain the backbone of most commercial CAD today.

Solid modeling: from surfaces to objects with insides

A surface model describes a skin. It cannot answer whether a point is inside the object, and that question is essential for manufacturing: you cannot reliably compute a volume, a mass, or a machining path from a skin alone. Solid modeling, developed mainly from the 1970s onward, set out to represent objects as closed, unambiguous solids. Two rival approaches emerged, and their competition shaped the field.

Constructive solid geometry (CSG) builds objects by combining simple primitives—blocks, cylinders, spheres—with Boolean operations: union, intersection, and difference3. A part is stored as a tree of operations, so a hole is literally "this block minus this cylinder." CSG is compact and robust in principle, because the primitives are exact, but extracting a surface mesh from a CSG tree for display or analysis requires evaluating the tree, which can be expensive and delicate at the seams.

Boundary representation (B-rep) instead stores the solid by its boundary: faces, the edges where faces meet, and the vertices where edges meet, with enough topological information to know which side is inside4. B-rep gives direct access to the surface geometry, which suits display and machining, but maintaining a consistent, watertight boundary through complex operations is hard—small numerical errors can leave a face unclosed or an edge shared by the wrong number of faces. The two representations coexisted and still do: many systems keep a CSG history alongside a B-rep result, and the long-running research problem of "non-manifold" modeling—representing shapes where more than two faces meet at an edge, or where a surface touches itself—grew out of the practical failures of both.

Parametric and variational design

As CAD spread through industry in the 1980s and 1990s, a further shift changed how designers worked. Instead of fixing geometry directly, parametric design lets a model be defined by dimensions and constraints—this hole is 10 mm from that edge, this face is parallel to that one—and the system solves for geometry that satisfies them. Variational design goes further, treating the constraints as an equation system to be solved when any dimension changes, so that editing one number ripples correctly through the whole model. This made models reusable as families of parts and made design intent explicit. The approach is closely associated with the rise of feature-based and history-based CAD in the late 1980s and 1990s, and it is now the default working style in mechanical design software. Its weakness is fragility: a constraint system can become over- or under-constrained, and a change that seems small can make the solver fail or produce a surprising shape.

Subdivision surfaces: smoothness from coarse control meshes

By the 1990s, animation studios faced a different problem. Character models needed to be smooth, but animators wanted to control them with relatively few, easily edited polygons. Subdivision surfaces answered this: start with a coarse polygon mesh and repeatedly refine it by a fixed rule, each step inserting new vertices and averaging positions so the mesh converges to a smooth limit surface. The Catmull–Clark scheme, published by Edwin Catmull and Jim Clark in 1978, generalizes B-spline surfaces to arbitrary meshes and produces smooth surfaces from quad-based control cages; the Loop scheme, published by Charles Loop in 1987, does the same for triangle meshes56. The practical breakthrough was that a modeler could work with a simple cage and get a smooth surface automatically, with the cage's shape controlling the result. Subdivision surfaces became standard in film production—Pixar's use of them in feature animation in the 1990s and 2000s is well documented—and they now sit alongside NURBS in CAD as well, since subdivision can represent shapes that are awkward for patch-based splines.

Implicit surfaces and distance fields

A parallel tradition describes shape not by a boundary but by a function: an implicit surface is the set of points where some function equals zero, with the sign of the function telling inside from outside. A distance field stores, for each point in space, the distance to the nearest surface, which makes it easy to blend shapes, offset them, and test proximity. These representations are compact for organic and blobby forms and are natural for simulation and collision. They became practical for graphics as volumetric grids and, later, sparse and adaptive data structures made them affordable, and they underpin techniques for modeling liquids, soft bodies, and scanned objects. Their weakness is resolution: a grid-based field must be fine enough to capture detail, and extracting a surface mesh from it (the marching cubes algorithm, published by Lorensen and Cline in 1987, is the classic method) produces many small triangles.

Geometry processing: making meshes usable

As scanned and modeled meshes became common in the 1990s and 2000s, a new set of problems appeared: real meshes are noisy, have irregular connectivity, and are often too dense or too sparse for a given task. Geometry processing grew as the discipline of cleaning, simplifying, smoothing, parameterizing, and analyzing meshes7. Mesh simplification reduces triangle count while preserving shape; remeshing rebuilds connectivity for better quality; parameterization flattens a curved surface onto a plane so textures can be painted; and spectral methods analyze a mesh's vibration-like modes to find features. This work turned the mesh from a mere display format into an object of study in its own right, and it connects modeling to rendering, simulation, and 3D printing.

Neural shape representation

The most recent turn borrows from machine learning. Neural shape representations encode a shape in the weights of a neural network—for example, a network that, given a 3D point, predicts whether it is inside the object or how far it is from the surface8. Such representations can be trained from examples, can interpolate between shapes, and can be queried at any resolution. They are young and still being evaluated against the classical representations, but they inherit the field's oldest question: what is the right way to describe shape so that a machine can store it, edit it, and reason about it? The answer today is plural—NURBS and B-rep in engineering, subdivision and polygon meshes in animation, implicit fields in simulation and scanning, and neural fields in research—and the open problems are the old ones in new dress: robustness of Boolean operations, guaranteed watertightness, smooth continuity across representations, and the conversion between them without loss.

References

  1. Geometric modeling, Wikipedia
  2. Computational geometry, Wikipedia
  3. Constructive solid geometry, Wikipedia
  4. Boundary representation, Wikipedia
  5. Loop subdivision surface, Wikipedia
  6. Catmull–Clark subdivision surface, Wikipedia
  7. Geometry processing, Wikipedia
  8. Neural field, Wikipedia

The frameworks of Geometric Modeling

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1960-1985

    Bézier and Coons Patch Modeling

    How can a few control points define a smooth free-form curve or patch?

    Fly to its tower in Nousfarer

  2. 1972-Present

    Non-uniform rational B-spline

    How can splines give local control and exact conics?

    Fly to its tower in Nousfarer

  3. 1972-Present

    Solid Modeling

    How can a computer represent an object with an inside?

    Fly to its tower in Nousfarer

  4. 1975-Present

    Boundary Representation

    How can a solid be stored by its bounding faces and edges?

    Fly to its tower in Nousfarer

  5. 1975-Present

    Constructive Solid Geometry

    How can solids be built by combining primitives with Boolean operations?

    Fly to its tower in Nousfarer

  6. 1978-Present

    Subdivision surface

    How can repeated refinement of a coarse mesh yield a smooth surface?

    Fly to its tower in Nousfarer

  7. 1982-Present

    Implicit Surfaces and Distance Fields

    How can a function's zero set and sign describe a shape?

    Fly to its tower in Nousfarer

  8. 1985-Present

    Parametric and Variational Design

    How can dimensions and constraints drive geometry automatically?

    Fly to its tower in Nousfarer

  9. 1995-Present

    Geometry Processing

    What algorithms operate on and improve digital shape meshes?

    Fly to its tower in Nousfarer

  10. 2018-Present

    Neural Shape Representation

    How can neural networks learn to represent and generate shape?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Bézier and Coons Patch ModelingNon-uniform rational B-splineSolid ModelingBoundary RepresentationConstructive Solid GeometrySubdivision surfaceImplicit Surfaces and Distance FieldsParametric and Variational DesignGeometry ProcessingNeural Shape Representation

Beyond the field

Where Geometric Modeling's ideas came from in other fields, and what they led to there: the strongest 12 of 27 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Geometric Modeling is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Geometric Modeling

Related fields

Fields its ideas are linked to

More in Computer Graphics

See where Geometric Modeling sits on the Map of Human Knowledge