PLAYING WITH PRISMS: WHAT OUR GPU FIRE DETECTOR REVEALED ABOUT GEMSTONE DISPERSION

If brightness is how much white light a gemstone returns to your eye, and scintillation is how its facets wink on and off as the stone moves, fire is the magic trick that makes a colorless crystal look like it is burning from within: the splitting of white studio light into vivid flashes of crimson, amber, emerald, cyan, and violet.

Throughout this post, every single studio gemstone and GPU diagnostic map we display is rendered as a strictly colorless crystal with zero body-color absorption. In a deeply colored ruby or emerald, the crystal's own body pigment acts like a color filter that swallows whole bands of the spectrum before they can reach your eye. Stripping away all body color leaves pure prism physics on stage: wherever you see a flash of red, gold, or electric blue in the stones below, you are looking directly at white light that has been pried apart by the angles of the cut and the dispersion of the mineral.

Rather than spending this article walking through shader code, we wanted to use our GPU Fire Detector as a laboratory instrument. Briefly, the detector places an uncolored crystal under discrete pinpoint lamps across nineteen viewing angles from -45 to +45 degrees, traces separate red, green, and blue rays through up to eight internal reflections, and measures how strongly the three color channels pull apart at each pixel. Once you can measure fire across hundreds of stones in a few seconds, you can run controlled experiments that expose some surprising truths about how materials, crown angles, flat tables, viewing tilt, and internal path length really work.

The Two-stage Prism And What The Gpu Fire Detector Sees#

Why do some colorless stones throw broad, saturated rainbow flames across the room while others return only crisp white and silver light?

A gemstone is not a single triangular prism from a high-school physics lab; it is a two-stage compound prism folded around a mirror-lined pavilion cone:

Look at the two stages on the left side of the diagram above:

1. Stage 1 (The Internal Optical Lever Arm): When a ray of white light enters an angled crown facet, blue light bends slightly more sharply than red light. Right at the surface, however, that initial angle split is tiny, often just a fraction of a degree. If those rays exited the stone immediately, red and blue would still overlap on the same spot of your retina and blend right back into white light. To become a visible rainbow flash, the split rays have to travel a physical distance inside the crystal, bouncing across the pavilion mirrors so that their tiny angular divergence grows into a wide spatial separation before they reach the top of the stone.
2. Stage 2 (Grazing Amplification at the Crown Exit): When those separated red, green, and blue rays hit an inclined crown facet on their way out, Snell's law bends them a second time. Here is the key physical nonlinearity: the closer an exiting ray hits to the crystal's critical angle, the more violently Snell's law stretches the angle between red and blue. Just before a ray is trapped by total internal reflection, the rainbow fan explodes in angular width.

In contrast, look at what happens when light enters and exits straight through a flat 0-degree table: because the top table is horizontal, rays traveling straight up and down hit at near-zero angle where Snell's law produces almost zero bending and zero color separation. On the right side of the diagram, you can see the actual GPU Fire Detector maps on four colorless cuts: a shallow windowed slab leaks its light out the back (0.19x fire), a classic Hearts and Arrows round brilliant wakes up around its rim (1.00x), a 3-fold diamond trilliant (Boreal Chartreuse Escapement) reaches 1.20x average fire and over 2.0x at tilt, and a 97-facet oval (Strobe Cassiopeia Oval) lights up across its entire face at more than double (2.32x) the fire of a classic diamond brilliant.

Experiment 1: The Material Paradox And Our Pipeline's Blind Spot#

Our first experiment started with a candid look at what our own software pipeline takes into account, and what it leaves out.

In real gemology, a transparent crystal has three distinct optical properties that govern how it bends and splits light:

• Refractive Index ($n$): How strongly the crystal slows down and bends light overall. This sets the critical angle of the pavilion mirrors and determines how much white light is trapped inside the stone versus leaking out the back.
• Chromatic Dispersion ($\Delta n$): How much more strongly the crystal bends violet and blue light than red light. This sets how wide the rainbow fan opens at every refraction.
• Double Refraction / Birefringence ($n_o$ and $n_e$): Only cubic crystals like Diamond (RI 2.42), Cubic Zirconia (RI 2.15), Spinel (RI 1.72), and Garnet are truly singly refractive (isotropic), bending light by one uniform refractive index in every direction. Most other gem minerals, including Moissanite (RI 2.648 to 2.691, birefringence 0.043), Zircon (RI 1.925 to 1.984, birefringence 0.059), Sphene (RI 1.885 to 2.050, birefringence 0.120), Sapphire (RI 1.762 to 1.770), and Quartz (RI 1.544 to 1.553), are doubly refractive (anisotropic). When a beam of light enters a doubly refractive crystal at an angle to its optic axis, the crystal splits that single beam into two orthogonally polarized rays (an ordinary ray and an extraordinary ray), each traveling at its own speed, bending at its own angle, and obeying its own critical angle.

Here is the confession about our own pipeline: originally, a stone in our faceting language and catalog generator only specified a single scalar refractive index, collapsing double refraction into one average number and omitting a separate dispersion constant. Even more importantly, because tracing three spectral wavelengths per pixel triples GPU raytracing time (and tracing six polarized-wavelength rays for birefringent crystals would multiply it by six), our standard catalog optimization pipeline runs as an isotropic single-RI raytracer with its fire weight set to zero, evaluating fire only once at the very end on the finished stone (unless you explicitly switch on the High-Fire profile in ProFacet Studio). Whenever a script omitted dispersion, our Fire Detector had to guess the material's dispersion from a lookup table based on that single refractive index alone, and then normalize every stone's raw fire against a single benchmark: a classic Hearts and Arrows diamond (RI 2.42, dispersion 0.044).

What happens when we test the exact same cut geometries across all eight catalog refractive indices, and compare our pipeline's single-number refractive-index assumption against real colorless and near-colorless minerals?

Four insights jump right out of the data and crystal physics above:

1. Why Cubic Zirconia (RI 2.15) Beats Diamond (RI 2.42) by Nearly 60% in Fire: Beginners often assume that because Diamond has a higher refractive index than Cubic Zirconia (2.42 vs. 2.15), Diamond must have more fire. In reality, refractive index and dispersion do not march in lockstep. Cubic Zirconia has 36% higher chromatic dispersion than Diamond (0.060 vs. 0.044), and Spinel (RI 1.72) has higher dispersion than Sapphire (RI 1.76). When you cut the exact same Hearts and Arrows diagram in colorless Cubic Zirconia and colorless Diamond, Cubic Zirconia produces 57% more rainbow fire (1.57x vs. 1.00x)!
2. When Two Minerals Share the Same Refractive Index, a Single Number Lies: Look at the right panel of the chart above. Whenever a shader guesses dispersion from refractive index alone, any two minerals that share a similar refractive index get assigned the same dispersion. Compare Colorless Diamond (RI 2.42) with Strontium Titanate (RI 2.41), the classic laboratory crystal famous for its dazzling rainbows. Because 2.41 sits right next to 2.42, a refractive-index lookup mistakes Strontium Titanate for Diamond and predicts 1.01x fire, whereas its true mineral dispersion (0.190 vs. 0.044) produces more than four times (4.37x) the fire of Diamond! The same blind spot hits natural Demantoid Garnet (RI 1.88) and Sphene (RI 1.95): both sit well below Diamond in refractive index, so a simple refractive-index curve scores them at 0.75x and 0.84x, yet both possess stronger dispersion than Diamond and out-fire it in reality (1.30x and 1.17x). Directly because of this finding, we upgraded our Faceting Scripting Language (FSL) and all of our procedural stone generators to declare an explicit dispersion property right next to ri in the source header (locking our reference Hearts and Arrows benchmark to 0.044 dispersion at RI 2.42, while allowing high-fire crystals like Strontium Titanate to specify 0.190 dispersion at RI 2.41).
3. Lower-RI Cuts Actually Have Higher Geometric Prism Efficiency Per Unit of Dispersion: Because our catalog normalizes every stone against Diamond without dividing out the mineral's dispersion constant, raw fire scores mostly reflect mineral chemistry (Moissanite at 0.104 dispersion naturally scores near 1.9x, while Quartz at 0.013 dispersion scores around 0.5x). What if we divide each stone's fire by its material dispersion relative to Diamond, isolating pure Geometric Prism Efficiency? Look at the bottom-left tiles in the diagram: per unit of material dispersion, a cut in Quartz (1.95x prism efficiency) or Sapphire (1.79x) converts dispersion into angular rainbow spread more than twice as efficiently as the same cut in Moissanite (0.83x)! Why? Because lower-RI crystals have a much larger critical angle (40.5 degrees in Quartz vs. 22.0 degrees in Moissanite) and much lower surface Fresnel glare, allowing internal rays to exit at steeper grazing angles right next to the critical-angle cliff where Snell amplification is strongest.
4. The Double-RI Wildcard (How Birefringence Splits One Rainbow into Twin Ghost Fans): What happens when a fiery crystal is also strongly doubly refractive, such as Moissanite (RI 2.648 to 2.691, birefringence 0.043), Colorless Zircon (RI 1.925 to 1.984, birefringence 0.059), Sphene (RI 1.885 to 2.050, birefringence 0.120), or Synthetic Rutile (RI 2.616 to 2.903, an astonishing 0.287 birefringence)? Before chromatic dispersion even finishes fanning red from violet, double refraction forks every entering beam into two separate polarized rays that bounce through the pavilion at slightly different angles, producing the classic "facet doubling" you see through a 10x loupe. Each of those two polarized rays then unfolds its own rainbow spectrum! Because our GPU pipeline averages double-RI minerals into a single isotropic index (for example, treating Moissanite as a single RI 2.67), it misses two real-world birefringent effects: first, when the ordinary and extraordinary rays exit cleanly side by side, a doubly refractive crystal can throw twin rainbow flashes from a single facet; second, if the rough is not oriented with its table perpendicular to the crystal's optic axis, those two offset rainbow fans can overlap out of phase (laying the red edge of one ray on top of the blue edge of its twin), softening crisp facet junctions and remixing part of the spectrum back into velvety white light.

Experiment 2: Steep Crowns, Flat Tables, And The Grazing-exit Trade-off#

For our second experiment, we held the material fixed (colorless crystal at RI 2.42) and swept the crown angle of a round brilliant from a pancake-flat 12 degrees all the way up to a steep 50 degrees, while also sweeping the width of the flat horizontal table from a 0% tableless dome to an 80% wide window.

Look at the left panel above. This single chart explains one of the oldest debates in diamond cutting:

• The Brightness & Sparkle Peak (24 to 26 Degrees): If all you want is maximum white light return and rapid pinpoint sparkle, a relatively shallow 24 to 26 degree crown with a generous table opens the widest possible window into the pavilion mirrors. Brightness reaches 1.21x and Scintillation V3 hits 1.63x, but dispersive fire lags behind (0.97x) because rays exit through gently sloped facets far from the critical angle.
• The Steep-Crown Prism Climb (36 to 50 Degrees): As you steepen the crown past 30 degrees toward 46 and 50 degrees, brightness falls by nearly 40% and sparkle drops by more than 60%, because the steep crown facets start bouncing internal rays back down into the pavilion via total internal reflection. Yet dispersive fire climbs relentlessly, jumping nearly 70% higher from 0.75x at 12 degrees to 1.13x at 46 degrees and 1.27x at 50 degrees! Every ray that manages to escape those steep upper bevels exits at a grazing angle right against the critical-angle cliff, fanning red and blue wide apart.
• The 28 to 36 Degree Sweet Spot: Classic brilliant cuts park their crown angles right between 30 and 35 degrees not because 34.5 degrees maximizes either brightness or fire alone, but because it sits right at the crossover where you keep over 90% of peak brightness while unlocking strong crown-bezel fire.

What about the flat table in the center of the crown? When we swept table width from 80% down to 11%, expanding a flat horizontal table past 65% crushed fire by 15% and sparkle by nearly half. You can see this difference immediately in the two colorless Moissanite (RI 2.67) heptagons below, both cut on the exact same 7-fold outline:

Experiment 3: The 19-angle Tilt Awakening (why Flat Tables Sleep At 0 Degrees)#

Now look at the right-hand chart of the previous diagram, where we plotted the GPU Fire Detector across all nineteen viewing tilt angles from -45 degrees to +45 degrees.

Have you ever noticed that when you stare straight down (0 degrees) into a classic round brilliant under an overhead spotlight, the center of the stone looks mostly bright white and silver, and then the moment you tilt your hand by 15 or 25 degrees, big flashes of crimson, orange, and electric blue suddenly ignite across the table?

Our 19-angle curves show that this is not an optical illusion; it is a real physical tilt awakening:

• In Classic Hearts and Arrows (RI 2.42): Dead-center at 0 degrees tilt, normalized fire is only 0.72x. Rock the stone to 20 or 30 degrees tilt, and fire surges to 1.44x, literally doubling as the stone tilts!
• In 3-Fold Trilliants (Boreal Chartreuse Escapement at RI 2.42 and Celadon Silhouette Triad at RI 2.15): Face-up fire at 0 degrees starts at 0.81x and 1.44x, then erupts under tilt to peaks of 2.02x and 2.17x!
• Why Tilting Turns a Flat Table into a Cross-Prism: When a stone sits level at 0 degrees, overhead light entering the flat horizontal table and exiting back through the flat horizontal table crosses two parallel surfaces (0 degrees in, 0 degrees out), which cancels dispersion just like looking through a flat windowpane. When you tilt the stone by 20 degrees, however, overhead light enters through the wide flat table at an oblique angle, bounces off the pavilion, and exits through the sloped crown bezel on the opposite side (or enters a sloped bezel and exits through the now-tilted table). Tilting the gem couples the wide aperture of the flat table with the steep angle of the opposite crown facets, turning the two surfaces into a giant asymmetric cross-prism!
• Why Domed Multi-Tier Ovals Fire Even at 0 Degrees: Look at the gold curve at the top of the chart for Strobe Cassiopeia Oval (RI 2.15, 97 facets). Unlike flat-tabled cuts that dip in the middle at 0 degrees, its 97-facet crown is already a terraced dome of angled prism facets right across the center, so it fires at 2.57x dead-center at 0 degrees and climbs to 2.92x under tilt.

Experiment 4: 526 Catalog Stones And The Internal Optical Lever Arm#

That brings us to our fourth and most surprising experiment. Remember that when our evolutionary optimizer generated the OffTheDop catalog, the weight for fire was set to zero. The optimizer only cared about brightness, spatial contrast, tonal entropy, sparkle, and avoiding windows.

So if the optimizer never tried to maximize fire, does every stone of the same material end up with roughly the same fire score? To find out, we ran 526 stones (all 40 curated Showcase and Signature benchmarks, 12 historical reference diagrams, and a stratified sample of 486 catalog stones across all eight materials and nine shape families) through the GPU Fire Detector:

Even with zero optimization weight on fire, cut geometry alone creates a two- to nearly five-fold spread in rainbow fire within the exact same material:

• Among Colorless Diamonds (RI 2.42), fire spans nearly a fivefold range, from 0.31x in shallow, flat-topped cuts up to 1.52x in deep multi-tier prisms!
• Among Cubic Zirconia (RI 2.15), fire spans a 3.4-fold range from 0.69x to 2.32x!
• Among White Sapphires (RI 1.76), fire spans a 3.0-fold range from 0.28x to 0.84x!

What geometric traits separate the highest-fire cuts from the lowest-fire cuts when the mineral is held constant? Looking at the correlation bars in the bottom-left panel above:

1. Crown Height (r = +0.51) and Total Depth (r = +0.49) Are the Twin Engines of Fire: Why do deeper stones with taller crowns generate so much more fire? Because of the Internal Optical Lever Arm. Deeper pavilions (60% to 75% total depth) and 3-fold or elongated outlines (trilliants, ovals, hexastars, and heptagons) force internal rays to travel a longer diagonal path across the belly of the crystal between reflections. That longer path gives the red and blue rays room to separate physically in space before they reach the crown, where a taller, steeper crown (16% to 19% crown height) bends them apart a second time.
2. Facet Subdivisions (r = +0.30) and Contrast Weave (r = +0.21) Help Catch the Split: Breaking the crown and pavilion into finer multi-tier facets (60 to 100 facets instead of 33) creates dozens of differently tilted exit windows, ensuring that at every viewing angle, a subset of virtual facets is sitting right in the grazing-exit sweet spot.
3. Fire Does Not Require Sacrificing Brightness (r = -0.19): While extreme 48 degree crown angles trade away brightness for fire, across well-proportioned catalog stones the correlation between geometric prism efficiency and brightness is only mildly negative (-0.19). By using a deeper internal lever arm and a multi-bevel crown instead of an overly steep single bezel, a cut can achieve both high brightness (1.05x to 1.34x) and exceptional fire (1.20x to 2.32x) at the same time!

Here are six standout Showcase and Signature Collection designs from our GPU fire study, rendered in our bidirectional path tracer as strictly colorless crystals so you can inspect their dispersive fire patterns directly:

Four Takeaways From The Fire Lab#

1. Always Separate Refractive Index, Dispersion, and Double Refraction: A mineral's refractive index, its chromatic dispersion, and its birefringence (double refraction) are independent physical properties. Cubic Zirconia (RI 2.15) out-fires Diamond (RI 2.42) by nearly 60% on the exact same cut; exotic crystals like Strontium Titanate (RI 2.41), Sphene (RI 1.95), and Demantoid Garnet (RI 1.88) throw far more fire than a simple refractive-index lookup predicts; and doubly refractive gems like Moissanite, Zircon, and Sphene split every ray into twin polarized rainbow fans that depend on crystal-axis orientation. When comparing cuts across different minerals, always normalize by the mineral's dispersion constant to see true Geometric Prism Efficiency.
2. Shrink the Flat Table and Add Angled Crown Bevels: A wide, flat 0-degree table is a rainbow dead zone when viewed straight on. Shrinking the flat table to 25% to 45% of the stone's width and surrounding it with two or three tiers of 28 to 38 degree crown bevels turns the upper dome into an active prism ring without choking off white brightness.
3. Design for the 20-Degree Tilt Cross-Prism: In flat-tabled cuts, half of the stone's fire only wakes up when the gem tilts 15 to 30 degrees in the wearer's hand, coupling the flat table with the opposite crown bezel as an asymmetric cross-prism.
4. Give the Rays an Internal Lever Arm: Angular color splitting at a single facet is useless unless red and blue rays have enough internal path length to separate in space before they exit. Slightly deeper pavilions (60% to 72% depth) and 3-fold trilliant, oval, or polygonal cross-bounce geometries act as optical lever arms that multiply rainbow fire while keeping face-up brightness high.