ONE STONE, EIGHT MASTERS: OPTIMIZING A PARAMETERIZED 57-FACET DIAMOND SRB FOR EVERY SINGLE METRIC

For more than a century since Marcel Tolkowsky published Diamond Design in 1919, the 57-facet Standard Round Brilliant (SRB) in diamond (refractive index 2.42, dispersion 0.044) has been the most scrutinized polyhedron in optics. Almost every discussion of diamond cutting treats the round brilliant as a single narrow target where pavilion mains hover near 40.8 degrees, lower-girdle halves reach roughly 75 to 80 percent of the pavilion radius, crown bezels sit near 34.5 degrees, and the flat octagonal table spans 53 to 59 percent of the girdle diameter.

What happens if we ask a radically different question? Suppose we take Tolkowsky's exact 57-facet round brilliant on a 96-tooth index gear, make every single facet tier and girdle height fully parameterized in FSL, lock the polyhedral topology with a strict assertion so the optimizer is never allowed to add or lose a single facet, and then optimize that stone from scratch for each optical metric in isolation:

1. Brightness only (total nineteen-tilt light return, ignoring contrast, sparkle, and fire).
2. Contrast (Combined) only (simultaneously maximizing spatial checkerboard frequency and bright-to-dark dynamic range).
3. Contrast Density only vs. Contrast Intensity only (splitting spatial contrast into its two constituent physical engines).
4. Scintillation only (dynamic nineteen-tilt flash and on-off sparkle).
5. Shannon Entropy only (thirty-two-bin tonal information richness across the stone's aperture).
6. Dispersive Fire only (three-wavelength RGB prismatic dispersion across all nineteen tilt angles).
7. Finally, All-Metric Optical Symphony (combining all optical metrics simultaneously to see whether a single 57-facet diamond can beat Tolkowsky 1919 across the board at once).

Every crystal in this study has the exact same 73 three-dimensional faces (57 optical crown and pavilion facets plus 16 vertical girdle facets), the exact same 96-index rotational indices, and the exact same diamond refractive index and dispersion. Yet as you will see below, chasing each optical metric to its mathematical limit morphs that single 57-facet topology into eight wildly different optical instruments.

In a traditional CAD drawing program, changing one pavilion or crown angle breaks the facet junctions unless you manually recalculate the decimal depths of every neighboring tier. Worse still, gemologists normally describe a round brilliant using ten or more numbers at once: pavilion main angle, lower-girdle angle, lower-girdle length percentage, girdle thickness, crown bezel angle, upper-girdle angle, star angle, star length percentage, crown height percentage, and table percentage.

Even more subtly, if you try to parameterize the pavilion by specifying raw angles for both the lower-girdle halves (P1) and the pavilion mains (P2), the two tiers constantly collide during optimization: whenever the optimizer tests a P2 angle steeper than P1, the P2 main slices away the P1 lower-girdle halves at the girdle rim and destroys the 57-facet topology!

When you write the 57-facet Standard Round Brilliant in FSL (Facet Scripting Language) using two-point edge interpolation (ep) and causal meetpoints (mp), both problems vanish at once: all ten gemological proportions collapse to six clean degrees of freedom, and the pavilion tiers can never cross out of order. Here is the complete, runnable parameterized FSL program that drives every experiment in this article (click Open in ProFacet Studio in the corner of the code block to run or edit it live):

name = "Parameterized 57-Facet Diamond SRB"
ri = 2.42
dispersion = 0.044
gear = 96

// Pavilion: 16 lower-girdle halves (P1) + 8 pavilion mains (P2) = 24 facets
P1 3 @ 42.00 +/- 14.0 : cp() xx 8
G1 3 @ 90.00 : 1.0000
P2 0 : mp(G1) : ep(edge(P1:3, P1:9), 0.223 +/- 0.777) x 8

// Girdle thickness: 16 vertical girdle facets
girdle = Point(0, 0, 0.030 +/- 0.050)

// Crown: 16 upper-girdle (C1) + 8 bezel mains (C2) + 8 stars (C3) + 1 table (T) = 33 facets
C1 3 @ 42.00 +/- 22.0 : mp(P1, P2) + girdle xx 8
C2 0 @ 34.50 +/- 22.0 : mp(G1, G1) x 8
C3 6 @ 21.50 +/- 20.0 : mp(C1) x 8
T  0 @ 0.00  : mp(C2) x 1

// Topological guardrails: lock 57 optical + 16 girdle = 73 faces & valid girdle
assert(stone().facetCount == 73)
assert(stone().isGirdleOK)

Let us trace how those eight cutting statements and two assertions parameterize the entire 57-facet diamond while preventing the optimizer from ever breaking the facet topology:

• Decoupling the Pavilion via P2 0 : mp(G1) : ep(edge(P1:3, P1:9), 0.223 +/- 0.777) x 8: First, P1 3 @ 42.00 +/- 14.0 : cp() xx 8 cuts the 16 lower-girdle halves to a sharp culet point at the bottom of the stone, and G1 3 @ 90.00 : 1.0000 cuts the 16-segment round girdle at unit radius. That leaves a crisp ridge edge(P1:3, P1:9) running from the culet apex (t = 0.0) all the way up to the girdle corner (t = 1.0). Instead of guessing an angle for P2, we define each pavilion main plane by two geometric targets: it must pass through the girdle meetpoint mp(G1) at index 0, and it must intersect the P1 ridge at fractional position t = 0.223 +/- 0.777 (clamped to (0.0, 1.0)). At t = 0.223, the lower-girdle halves reach exactly 1 - 0.223 = 77.7% of the way from the girdle toward the culet, and the emergent P2 pavilion main angle is 40.83 degrees! Better still, for any P1 angle between 28 and 56 degrees and any edge ratio t strictly between 0.0 and 1.0, the P2 main is mathematically guaranteed to land between the culet and the girdle without ever erasing P1.
• Crown Meetpoints Automatically Control Table Diameter and Star Length: On the crown, C1 starts right at the pavilion-girdle junction shifted up by the girdle vector (mp(P1, P2) + girdle), while the eight C2 bezel mains cut to mp(G1, G1). The eight C3 star facets then slice down to the inner apex of the upper-girdle pairs (mp(C1)), and the flat T table cuts at zero degrees to the inner apex of the bezel kites (mp(C2)). Consequently, steepening or shallowing C3 relative to C2 and C1 continuously expands or contracts the flat table from nearly 60 percent of the stone's width down to a 3.6 percent pinhole without ever breaking a meetpoint!
• Why assert(stone().facetCount == 73) Is an Unbreakable Topological Lock: In our constructive solid geometry engine, every facet is an infinite half-space plane slicing a convex 3D polyhedron. In a valid 57-facet Standard Round Brilliant with a 16-segment round girdle, the finished polyhedron has 33 crown faces + 24 pavilion faces + 16 girdle faces = 73 total faces. If the optimizer tries an illegal combination of crown angles, such as making C2 steeper than C1 (C2 >= C1) or making C3 so steep that it undercuts the entire table T, the steeper plane completely slices away the neighboring tier and causes stone().facetCount to drop below 73! Pairing assert(stone().facetCount == 73) with assert(stone().isGirdleOK) creates a razor-sharp boundary around the feasible 57-facet SRB polytope: any candidate that loses even a single facet or pinches the girdle waist is rejected in microseconds before a single light ray is traced.

When you optimize a six-dimensional gemstone inside a non-convex polyhedral boundary, a single local optimization run from Tolkowsky's 1919 starting point is not enough. Why? Because the optical landscape of a diamond contains multiple distinct physical regimes separated by total internal reflection cliffs: a shallow-crown single-reflection basin, a steep-crown double-refraction basin, and a high-depth multi-bounce pavilion W-path basin above 54 degrees.

To make sure we found the true global champion for every metric rather than a local foothills solution, we built a four-stage GPU optimization pipeline and ran it exhaustively across all eight objectives:

1. Global 6D Feasible Polytope Atlas (1,585 Feasible Geometries): Before launching any evolutionary search, we swept the full six-dimensional parameter box across P1 angles from 28 to 56 degrees, P2 edge ratios from 0.005 to 0.995, crown angles from 12 to 64 degrees, star angles, and girdle heights, filtering for stone().facetCount == 73 and stone().isGirdleOK and evaluating all nineteen tilt angles on the GPU for every surviving geometry.
2. Six Multi-Basin Differential Evolution Restarts per Objective: For each of the eight objectives, we seeded six independent SHADE Differential Evolution populations (five anchored in the top distinct geometric basins discovered in the global atlas plus one anchored at Tolkowsky 1919) and evolved each population across 55 generations on the GPU.
3. Warm-Start Evolutionary Refinement: We took the winning genome from the six restarts and ran two focused warm-start evolutionary refinement passes around the champion basin.
4. Five-Stage Multi-Scale Coordinate and Diagonal Pattern Search: Finally, we locked onto the exact peak by running a deterministic coordinate and coupled-diagonal pattern search on the GPU with step sizes stepping down to 0.01 degrees on all four angles, 0.001 on the P2 edge interpolation parameter, and 0.0001 on girdle height until no single-parameter or coupled move could improve the score by even one part in a million.

Look at what happened to the 57-facet round brilliant across those eight optimization runs when we plot every champion's exact top-down crown wireframe and side-profile silhouette at a common horizontal scale (with Tolkowsky's 1919 profile shown as a dashed reference outline behind each stone):

Let us begin with the two metrics at the heart of every cut-grading debate: white light return (Brightness) and spatial pattern contrast. Below are the GPU bidirectional path-traced studio renders of our Tolkowsky 1919 baseline (#0), the Brightness Only champion (#1), and the Contrast Combined champion (#2) (click Open in Studio in any caption to inspect and rotate the 3D stone live):

Look at the dramatic fork in the road between #1 Brightness Only and #2 Contrast Combined:

• Why the Brightness Champion Flattens the Crown to 18.3 Degrees: Notice that on the pavilion, the Brightness Only optimizer barely touches Tolkowsky's classic angles: it sets P1 to 41.87 degrees and pulls ep down from 0.223 to 0.105 (extending the lower-girdle halves to 89.5 percent and setting the emergent P2 angle to 41.07 degrees), because a 41-degree diamond cone is already a near-perfect two-bounce retroreflector. The shock happens on the crown: the optimizer pulls the bezel angle C2 all the way down from 34.5 degrees to 18.26 degrees, drops the upper-girdle C1 to 20.04 degrees, and shrinks the flat table to 29.8 percent! Why? Because in diamond, any ray rising from the pavilion that strikes a steep 34.5-degree crown facet at more than 24.4 degrees from the facet normal suffers a third total internal reflection back down into the pavilion instead of escaping to the eye! Flattening the crown dome to 18 to 20 degrees aligns the crown facets almost perpendicular to the rising pavilion rays, opening a huge escape hatch across the entire upper hemisphere and driving nineteen-angle brightness up to 160.5 percent of the Hearts and Arrows reference (45 percent brighter than Tolkowsky!). The price of that shallow crown is immediately visible in the metrics: dispersive rainbow fire drops by a quarter (down to 82.5 percent) because low-angle crown facets act as weak prisms.
• Why Contrast Combined Raises a Steep 44.4-Degree Crown Fortress: Now look at #2 Contrast Combined on the right. When you ask the optimizer to maximize spatial contrast (the combination of fine checkerboard edge density and local Michelson luminance cliffs across nineteen tilt angles) without caring about total brightness, it does the exact opposite of the Brightness stone: it pushes the crown bezel C2 up to 44.38 degrees, C1 up to 47.87 degrees, and stretches the lower-girdle halves to 93.1 percent (ep = 0.069), increasing total stone depth to 71.4 percent. Steep crown facets act as strong angular magnifiers and refractive prisms: adjacent virtual facets sample wildly different elevation bands of the lighting dome, lifting combined contrast to 110.0 percent and delivering a massive 146.0 percent fire bonus as a free side effect!

What happens when we decompose spatial contrast into its two separate components (Contrast Density versus Contrast Intensity) or optimize purely for dynamic Scintillation? Here are champions #3, #4, and #5:

Compare #3 Contrast Density Only and #4 Contrast Intensity Only: across our 1,585-stone global atlas, their pairwise correlation is only r = 0.27, because they pull the diamond in opposite physical directions! - To maximize Contrast Density (how finely the stone's aperture is woven with light-to-dark transitions), stone #3 drives the pavilion all the way up to the 56.00-degree P1 / 55.35-degree P2 multi-bounce cone with a 94.0 percent lower-girdle length (ep = 0.060) beneath a classic 35.08-degree crown bezel. Inside a 55-degree pavilion with 94 percent lower-girdle halves, every ray undergoes three to four internal reflections before exiting, multiplying the virtual facet count into a dense, intricate pinwheel of fine dark and bright wedges (113.7 percent Contrast Density and 151.9 percent Fire!). - To maximize Contrast Intensity (the sheer dynamic range between the brightest and darkest adjacent blocks), stone #4 stays on a 40.06-degree pavilion main with a shorter 85.6 percent lower-girdle split (ep = 0.144), raises the crown bezel C2 to 41.24 degrees, and drops the star facets C3 down to a nearly flat 6.63 degrees (shrinking the table to 24.5 percent). That huge 34.6-degree cliff between the steep bezels and flat stars coalesces the internal reflections into bold, high-amplitude black-and-white blocks (114.6 percent Contrast Intensity), at the cost of coarser spatial density (91.2 percent). - Meanwhile, #5 Scintillation Only reveals what drives pure dynamic sparkle when a diamond tilts: it pairs a 39.92-degree P1 / 39.20-degree P2 pavilion (ep = 0.076, giving a long 92.4 percent lower-girdle split) with a moderately low 25.70-degree crown bezel and a 10.45-degree star. Because the 92.4 percent lower-girdle halves slice the pavilion into 24 narrow mirror strips and the 25.7-degree crown lets rays escape cleanly across a wide angular cone, tiny five-degree tilts trigger rapid on-off flashes across the entire face-up stone, reaching 223.4 percent normalized scintillation (+68 percent over Tolkowsky!) while maintaining 141.8 percent brightness and 111.1 percent fire.

What happens when we optimize for Shannon Information Entropy (how evenly the stone populates all 32 luminance histogram bins from deep shadow through mid-tones to specular highlights), Dispersive Rainbow Fire, or All-Metric Optical Symphony? Here are champions #6, #7, and #8:

Every single one of the three stones above uncovers a fundamental piece of crystal physics:

• #6 Shannon Entropy Only (The 32-Shade Tonal Sculptor): A stone that only produces binary black and white blocks hits a ceiling on Shannon entropy because all thirty intermediate grey bins sit empty. To fill every luminance bin across nineteen tilts, the Entropy optimizer keeps Tolkowsky's exact 77.2 percent lower-girdle ratio (ep = 0.228), steepens the pavilion to P1 = 45.08 deg / P2 = 43.89 deg, thickens the girdle to 3.8 percent, and staggers the crown into three gentle steps (C1 = 20.00 deg, C2 = 16.63 deg, C3 = 13.30 deg). That angular staircase forces every tier of virtual facets to sample a different latitude of the upper hemisphere, pushing Shannon entropy to 107.3 percent while simultaneously lifting brightness to 118.6 percent and Contrast Density to 109.1 percent!
• #7 Dispersive Fire Only (The 3.6%-Table Prism Tower): This is arguably the wildest legal polyhedron in the entire experiment. Remember that assert(stone().facetCount == 73) forbids the optimizer from deleting the flat 0-degree table facet T, even though a flat 0-degree window has zero prism angle face-up! How does the evolutionary optimizer outsmart that constraint? It steepens C1 to 46.50 degrees, C2 to 45.78 degrees, and C3 to 38.83 degrees so that the crown rises into a tall 46-degree prism pyramid (crown height 49.5 percent!) capped at its very peak by a microscopic 3.6 percent octagonal table that legally preserves the 73rd face! Simultaneously, it leaps onto the 54.75-degree P1 / 54.12-degree P2 steep pavilion island (ep = 0.049, lower-girdle length 95.1 percent, total stone depth 120.7 percent). Inside that deep cone, rays undergo multi-bounce W-path reflections that double the optical path length and exit through the 46-degree crown prisms right near the critical-angle grazing boundary where Snell's law magnifies angular dispersion (d theta / d n) to infinity. The result is 247.8 percent normalized fire (more than 2.26 times Tolkowsky's fire, peaking at 0.146 raw dispersion at 15 degrees of tilt!), at the cost of sacrificing two-thirds of its white brightness (30.8 percent).

Now let us step back and look at the complete 9-stone by 8-column trade-off matrix below. Every row is one of our nine 57-facet Diamond SRBs; green cells beat Tolkowsky's 1919 baseline, red cells fall behind Tolkowsky, and gold borders highlight the exact metric each stone was optimized to win:

Look across Rows 1 through 7 of that matrix and you see Goodhart's Law written in solid diamond: whenever you elevate a single optical metric to be the sole target of optimization, the optimizer ruthlessly exploits the other metrics as sacrificial fuel:

• Optimizing Brightness Only (#1) wins a 160.5 percent light-return record, but sacrifices a quarter of its rainbow fire (82.5 percent) and flattens the crown into a 29.8 percent-table dome.
• Optimizing Contrast Combined (#2) and Contrast Intensity (#4) drives crown bezel angles above 41 to 44 degrees, boosting dark-arrow contrast while pulling white brightness down to 94.0 percent and 74.2 percent.
• Optimizing Fire Only (#7) more than doubles Tolkowsky's dispersion (247.8 percent), but crushes brightness down to 30.8 percent and stretches total depth to 120.7 percent.

Now look at Row #8 All-Metric Optical Symphony (P1 = 42.19 deg, ep = 0.077 giving P2 = 41.46 deg and 92.3% lower-girdle length, girdle = 0.0388, C1 = 41.38 deg, C2 = 37.33 deg, C3 = 19.41 deg, Table = 51.2%, Depth = 64.7%), shown in the right panel of the gallery above:

1. How ep = 0.077 (92.3% Lower-Girdle Halves) Unlocks Simultaneous Sparkle and Contrast Density: In Tolkowsky's 1919 design (ep = 0.223, 77.7% lower-girdle length), the eight pavilion mains occupy a wide share of the pavilion cone. By pulling ep down to 0.077 beneath a 42.19-degree P1, the optimizer lengthens the 16 lower-girdle halves to 92.3 percent and tightens the emergent P2 main angle to 41.46 degrees. Those longer, steeper lower-girdle mirrors subdivide the face-up reflection map into finer spatial checkerboards (111.7 percent Contrast Density vs. Tolkowsky's 102.2 percent) and boost dynamic tilt sparkle to 157.4 percent Scintillation (+18.5 percent over Tolkowsky!).
2. How the Two-Tier Crown (C2 = 37.33 deg, C3 = 19.41 deg) Lifts Both Brightness and Fire at Once: Across the random 6D atlas, Brightness and Fire are nearly orthogonal (r = 0.01), because steep crowns create fire while shallow crowns transmit white light. Stone #8 gets the best of both worlds by splitting the crown job across two tiers: it nudges the bezel mains C2 up by nearly three degrees to 37.33 degrees (shrinking the table to 51.2 percent so the outer crown ring acts as a high-dispersion prism, lifting Fire to 123.4 percent vs. Tolkowsky's 109.4 percent), while simultaneously shallowing the inner star facets C3 to 19.41 degrees so rising central rays escape without total internal reflection (lifting Brightness to 113.5 percent vs. Tolkowsky's 110.6 percent, Combined Contrast to 103.4 percent, and Shannon Entropy to 103.1 percent!).

After more than a century of service, Marcel Tolkowsky's 1919 equations still stand remarkably close to the multi-objective Pareto frontier of diamond. Yet when you parameterize all six degrees of freedom in FSL, guard the 73-face topology with assert(stone().facetCount == 73), and let a GPU explore the full nineteen-angle optical landscape, you can both map the extreme outer boundaries of crystal physics and tune Brightness, Contrast, Scintillation, Entropy, and Fire higher at the exact same time.

1. Tolkowsky, M. (1919). Diamond Design: A Study of the Reflection and Refraction of Light in a Diamond. E. & F. N. Spon, London.
2. Moses, T. M., Johnson, M. L., Green, B., Blodgett, T., Cino, K., Geurts, R. H., Gilbertson, A. M., Hemphill, T. S., King, J. M., Kornylak, L., Reinitz, I. M., and Shigley, J. E. (2004). "A Foundation for Grading the Overall Cut Quality of Round Brilliant Cut Diamonds." Gems & Gemology, Vol. 40, No. 3, pp. 202 to 228.
3. Cutler, B., Dorsey, J., McMillan, L., Müller, M., and Jagnow, R. (2002). "A Procedural Approach to Authoring Solid Models." ACM Transactions on Graphics (SIGGRAPH 2002), Vol. 21, No. 3, pp. 302 to 311.
4. Tanabe, R., and Fukunaga, A. (2013). "Success-History Based Parameter Adaptation for Differential Evolution." 2013 IEEE Congress on Evolutionary Computation (CEC), pp. 71 to 78.