Walk inside an optical system

Why can one object appear many times?

At each glass–air interface, Fresnel reflection creates a reflected route alongside the transmitted route. From inside glass, angles above the critical angle produce total internal reflection: θc(λ) = asin(1/n(λ)), about 41° for n = 1.52 in air. Several planar faces can therefore show repeated virtual images of room edges and emitters.

Unfolding each reflection into a mirrored copy of the room makes the paths straight. A rectangular mirror room produces a lattice of virtual rooms; a prism produces a pattern determined by its face angles. Glass also transmits light, so it is only mirror-like for some directions. Finite bounce limits truncate this pattern.

The grid in the supplied screenshots is not, by itself, proof of this effect: editor lines and geometry mismatches must also be excluded. The original rotated-box scene had enlarged analytical half-extents even though rotation was already applied separately; this mismatch with the visible mesh is corrected. Disable overlays when interpreting the rendered image. These geometric-optics lines are not diffraction or holographic fringes.

Plane mirrors and virtual images · OpenStax

Four controlled scenes

129 · Inside glass: start inside a large glass block, with asymmetric colored markers outside. Rotate to compare transmitted views and internally reflected images. This is an optically immersed camera probe; an air cavity around a real camera is not modeled.

130 · Mirror room: six reflective walls and asymmetric luminous markers reveal repeated rooms. Reflectance is 0.96, so long paths fade. A 32-surface-bounce budget keeps the experiment finite.

131 · Prism bench: a narrow broadband white source illuminates a triangular glass extrusion and a white screen. The three side faces refract light; opaque end plates close the fixture. Camera-connection light tracing follows light through glass onto the wall.

132 · Prism bench with mist: the same setup adds dilute homogeneous scattering. The fog reveals the paths; it does not create the wavelength separation.

A laser is not automatically a rainbow source

A monochromatic laser remains one color. Here the source is a narrow white-light beam. Wavelength-dependent refractive index separates its colors at nonparallel faces. A parallel-sided glass slab generally leaves the emerging direction parallel to the incident direction, with wavelength-dependent lateral displacement instead.

The original scene uses a rectangular prism and exaggerated dispersion (spread 12). The new bench uses a triangular cross-section and spread 1. They answer different questions: interesting reflected views versus a simple spectrum on a screen. A very broad source, excessive angular spread or a short screen distance can make the colors overlap again.

Dispersion by prisms · The Physics Classroom

Implementation and interpretation

Naive and camera-connection light tracing now sample a cone source directly within its angular support, using p(ω) = 1/[2π(1 − cos α)]. Throughput retains the surface cosine and divides by this density. Previously, a 0.2° cone accepted only about 0.0012% of cosine-weighted launch directions. This change improves sampling efficiency; it does not widen the source or invent extra light. BDPT retains its existing launch proposal and density evaluation.

The physical-dispersion wall spot is deliberately small. Walk closer to the screen to inspect it; the scene is not an exaggerated full-wall rainbow. The fog scene exposes the incident and outgoing beam paths. The finite render below remains noisy and is not a convergence reference.