implicit surfaces

Hart’s paper is the clearest foundational account of why sphere tracing works at all. It does not present marching as a heuristic loop, but as a geometric method built on distance bounds: if the field gives a trustworthy lower bound on distance to the surface, each step is guaranteed not to cross the zero set. The paper also frames this against the weaknesses of blind sampling and derivative-heavy root-finding methods, which is useful if you want to explain why sphere tracing became the robust default for many implicit scenes.

It is also stronger than most practical tutorials on the mathematical side. Hart ties the method to Lipschitz conditions, explains why a bound on derivative magnitude is enough even when derivatives are discontinuous or undefined, and connects the technique to antialiasing through cone-tracing-style reasoning. On top of that, the appendices broaden the paper from a rendering method into a modeling reference by deriving distance functions and operations for many important implicit primitives and deformations.

Bloomenthal’s survey situates ray marching inside the wider language of implicit modeling instead of treating SDFs as an isolated graphics trick. It gives a careful explanation of implicit functions, regularity, and the geometric meaning of the gradient, including why the normalized gradient provides a surface normal and how numerical differencing approximates that gradient when an analytic derivative is unavailable. That makes it one of the strongest references for the conceptual link between scalar fields, isosurfaces, and shading.

It is also particularly useful because it connects rendering to modeling operations. The survey covers constructive solid geometry, blending, and the broader convenience of implicit representations for boolean composition and shape design, including cases where hard-edged junctions challenge polygonal approximations. For a research page, that breadth matters: it helps explain why SDF workflows are appealing not only because they ray march well, but because implicit surfaces are naturally compositional and support operations that are awkward in mesh-first representations.