The Two Ways to Describe a Shape

Every CAD system needs a way to represent three-dimensional objects in memory. For the past four decades, the dominant answer has been boundary representation (B-rep): you store the surfaces, edges, and vertices that form the outer shell of a solid. This is how CATIA, SolidWorks, NX, and virtually every major commercial CAD kernel works.

There is another way. Instead of describing where the boundary is, you describe how far away any point in space is from the boundary. This is implicit geometry, and it is built on a mathematical construct called a signed distance field (SDF).

Signed Distance Fields Explained

A signed distance field is a scalar function defined over three-dimensional space. For every point (x, y, z), the function returns a single number:

Negative values mean the point is inside the solid.
Positive values mean the point is outside the solid.
Zero means the point lies exactly on the surface.

Think of it like a weather map for geometry. A topographic map assigns an elevation to every point on a 2D plane. An SDF assigns a distance-to-surface to every point in 3D space. The surface itself is the “sea level” contour where the value crosses zero.

A Simple Example

Consider a sphere centered at the origin with radius 5. Its SDF is:

f(x, y, z) = sqrt(x^2 + y^2 + z^2) - 5

At the center (0, 0, 0), the value is -5, deep inside. At (5, 0, 0), the value is 0, exactly on the surface. At (10, 0, 0), the value is 5, outside. No list of triangles, no NURBS patches. Just a function.

Why This Matters for Engineering

The mathematical elegance of SDFs is not merely academic. It has concrete engineering consequences.

Boolean Operations Become Trivial

In B-rep, computing the intersection of two solids is one of the hardest problems in computational geometry. Entire PhD theses have been written about robust Boolean operations on NURBS surfaces. Edge cases involving tangent surfaces, coincident faces, and degenerate sliver geometry can crash commercial kernels.

With SDFs, a Boolean union of two shapes A and B is simply:

union(A, B) = min(A(p), B(p))

Intersection is max(A, B). Difference is max(A, -B). These operations are numerically stable, trivially parallelizable, and work regardless of geometric complexity.

Smooth Blending Is Native

B-rep systems require dedicated fillet and chamfer operations that modify the topology of the boundary. These operations frequently fail on complex geometry, especially when fillet radii conflict or surfaces cannot be extended.

In implicit geometry, smooth blending is a modification of the distance function itself. A smooth union operator replaces the sharp min with a differentiable approximation, producing a rounded blend whose radius is controlled by a single parameter. No topology changes. No surface trimming. No failure modes from conflicting radii.

Lattice Structures and Infill

Implicit geometry excels at representing structures that are essentially impossible in B-rep: triply periodic minimal surfaces, stochastic lattices, and functionally graded infill patterns. These structures are critical for additive manufacturing, lightweight aerospace components, and biomedical implants.

An SDF for a gyroid lattice, for example, is a compact trigonometric expression. The same structure described as B-rep would require millions of individual surface patches.

Per-Point Material Properties

Because an SDF is a field defined over space rather than a boundary, it is natural to define other fields over the same domain: material density, thermal conductivity, Young’s modulus. This enables multi-material design workflows where material assignment varies continuously through the volume, not just at discrete part boundaries.

How Implicit Geometry Compares to B-rep

Neither representation is universally superior. Each has domains where it dominates.

B-rep strengths:

Mature ecosystem with decades of tooling
Exact representation of canonical surfaces (planes, cylinders, cones, spheres)
Well-understood tolerancing and GD&T workflows
Direct compatibility with CNC machining toolpath generation

Implicit geometry strengths:

Robust Boolean operations with no topological failure modes
Native smooth blending and filleting
Compact representation of lattice and infill structures
Natural support for multi-material and functionally graded design
Amenable to GPU parallelization and real-time evaluation

The most powerful modern systems will likely support both representations and provide bridges between them. A designer might model the outer shell of a part using traditional parametric features, then fill the interior with an SDF-defined lattice, and finally export the combined result for 3D printing.

Real-World Analogies

If you have ever used a metal detector on a beach, you have experienced something like an implicit field. The detector does not know the shape of the buried object. It only knows the signal strength at each point as you sweep. Strong signal means you are close. Weak signal means you are far. The object’s “surface” is the contour where the signal crosses a threshold. You reconstruct the boundary from the field, not the other way around.

Another analogy: MRI scans. A magnetic resonance image is a scalar field of tissue density. The boundary of an organ is not stored explicitly. It is the contour where density changes. Radiologists extract surfaces from fields every day. Implicit CAD does the same thing for engineering geometry.

Where NeuroCAD Fits

NeuroCAD is built from the ground up on implicit geometry. Its core evaluation engine computes signed distance fields across a distributed constraint graph, using e-graph optimization to simplify and accelerate field expressions. This architecture means that operations like Boolean combination, smooth blending, and lattice generation are first-class primitives rather than afterthought plugins.

The kernel exposes a graph-based design history where every node is a field operation. This makes deterministic replay, undo/redo, and collaborative editing structurally simple, because the design is a directed acyclic graph of field transformations rather than a fragile sequence of boundary mutations.

NeuroCAD does not discard B-rep entirely. Surface realization modules can extract boundary meshes from the implicit field when needed for simulation, rendering, or export to legacy formats. But the source of truth remains the field, not the boundary.

The Future of Geometry in CAD

Implicit geometry is not new. Researchers have published on SDFs and constructive solid geometry with implicit primitives since the 1970s. What is new is the hardware to make it practical. Modern GPUs can evaluate billions of field samples per second. WebGPU brings that capability to the browser. Additive manufacturing has created demand for geometry types that B-rep handles poorly.

The convergence of these trends means that implicit geometry is moving from research curiosity to production capability. Engineers who understand SDFs will have a meaningful advantage as the tools mature and the design possibilities expand.

Summary

Implicit geometry represents shapes as scalar fields over space rather than as explicit boundary surfaces. Signed distance fields are the most common form. This approach makes Boolean operations trivial, smooth blending native, lattice structures compact, and multi-material design natural. It complements rather than replaces B-rep, and modern CAD platforms increasingly need both representations to serve the full range of engineering needs.

What Is Implicit Geometry in CAD?