Settings›Optics

Optics

Refraction, reflection and dispersion share a screen-space dielectric model.

Index and optical path

ior is the refractive index: 1–2.4, default 1.5. An index of 1 matches air and causes no geometric refraction. Snell refraction uses a thickness-bounded curved profile. thickness is optical depth in CSS pixels, 0–96, default 6; depth is curve width. The independent strokeThickness, 0–8 px, default 1, controls the rim line.

Fresnel reflection

When reflection is enabled, the unpolarized dielectric Fresnel equations mix transmitted and reflected colors. Reflection strengthens at grazing surface angles. The environment is the supplied scene image; hidden 3D surroundings are not reconstructed.

Two dispersion models

dispersion="0" disables channel separation. Physical separation follows surface incidence and optical path, not the decorative directional light angle. It is not calibrated to a glass catalog or a full spectral solution.

<edinstvo-glass ior="1.5" thickness="12" depth="12"
  reflection="1" dispersion-mode="physical" dispersion="0.5" frost="0.1">
  Content
</edinstvo-glass>

Limits and rendering checks

This remains a 2D lens under an orthographic view. No repeated internal reflections, polarization or full spectral transport is simulated. Tint, lift, bloom, touch warping and contrast adaptation are appearance controls; the combined material is not a fully energy-conserving spectral model.

Browser tests read rendered contour, normal and displacement buffers. They check symmetric gap-free curves, continuity at merged bridges, deep capsules and zero geometric refraction at IOR 1, at DPR 1 and 2. They protect specific regressions rather than proving every parameter combination artifact-free.

Performance measurements

The final local fixture uses 16 surfaces, 40 steps, viewport 1100×850 and DPR 1. Warm-frame texture allocations, uniform lookups and idle draws are zero. Median step duration is 42.6 ms, p95 120.1 ms; a step waits for two animation frames, so these are not FPS figures. The former eight-composite cutoff was removed because it lost lower glass layers. Correct composition costs more work; no overall FPS gain is claimed.

Blur targets are reused. The source LRU targets 64 MiB and 32 entries, allowing an active source to exceed the budget alone. SVG atlases have a separate 32 MiB target. Full 17-sample Gaussian blur is retained because paired sampling on fractional mip levels produced visible Retina stripes.