ellipsoidSDF: Three-Dimensional Ellipsoidal Distance Field

Non-Uniform Spherical Geometry with Axis-Independent Scaling

The ellipsoidSDF function generates signed distance fields for ellipsoidal shapes by applying independent scaling factors along each coordinate axis. This primitive extends spherical geometry to accommodate stretched or compressed forms in any direction.

Mathematical Foundation

The ellipsoid distance function operates through normalized coordinate transformation:

$$k_0 = \left|\frac{\vec{p}}{\vec{r}}\right|$$ $$k_1 = \left|\frac{\vec{p}}{\vec{r}^2}\right|$$ $$d_{\text{ellipsoid}} = \frac{k_0(k_0 - 1)}{k_1}$$

where $\vec{r} = (r_x, r_y, r_z)$ defines the semi-axis lengths along each coordinate direction.

The function approximates the exact distance through iterative refinement, providing accuracy suitable for raymarching applications while maintaining computational efficiency.

Function Signature

Parameter	Type	Description
p	vec3	Sample point position
r	vec3	Semi-axis radii (x, y, z scaling factors)

Implementation Demonstrations

ライブエディター

const fragment = () => {
      const up = vec3(0, 1, 0)
      const eps = vec3(0.01, 0, 0)
      const eye = rotate3dY(iTime).mul(vec3(5))
      const args = [vec3(1.5, 0.8, 1.2)]
      const march = Fn(([eye, dir]: [Vec3, Vec3]) => {
              const p = eye.toVar()
              const d = ellipsoidSDF(p, ...args).toVar()
              Loop(16, ({ i }) => {
                      If(d.lessThanEqual(eps.x), () => {
                              const dx = ellipsoidSDF(p.add(eps.xyy), ...args).sub(d)
                              const dy = ellipsoidSDF(p.add(eps.yxy), ...args).sub(d)
                              const dz = ellipsoidSDF(p.add(eps.yyx), ...args).sub(d)
                              return vec4(vec3(dx, dy, dz).normalize().mul(0.5).add(0.5), 1)
                      })
                      p.addAssign(d.mul(dir))
                      d.assign(ellipsoidSDF(p, ...args))
              })
              return vec4(0)
      })
      const z = eye.negate().normalize()
      const x = z.cross(up)
      const y = x.cross(z)
      const scr = vec3(uv.sub(0.5), 2)
      const dir = mat3(x, y, z).mul(scr).normalize()
      return march(eye, dir)
}