a jill × tim collaboration · apr 2026

What if a bar ordered
The Gates of Hell?

A drinking establishment in need of a door hires an artist who knew the meaning of greatness. You run Doors-R-Us. You have a laser cutter, a CNC mill, a bronze foundry, and 96 hours. This scenario says no.

Model boundary: this is an illustrative scenario, not a fabrication quote, engineering analysis, or verified business plan. The timing, cost, optical, and machining assumptions require independent domain validation.

0 mheight
0 mwidth
0figures
0 yrsRodin worked
0 hrsuntil Tuesday

This is the door.

A local still from a credited photogrammetry scan of the Stanford cast. The external interactive viewer is linked below rather than embedded.

RECONSTRUCTING 0 / 175,700 triangles
Local still of the Gates of Hell photogrammetry model
Open Alan Zimmerman's external model · CC BY 4.0 · Stanford, Cantor Sculpture Garden
I. The commission

Thirty-seven years of overtime.

August 16, 1880. Edmond Turquet, Under Secretary of Fine Arts, hands Auguste Rodin 8,000 francs and a brief: bronze doors for a planned Musée des Arts Décoratifs, to be built on the ruins of the Cour des Comptes — the court of auditors that burned during the Paris Commune of 1871. Deadline: 1885.

Rodin had just returned from Italy, where he stood in front of Ghiberti's Gates of Paradise and understood that a door could be a masterpiece. He chose Dante. "For a whole year I lived with Dante, with him alone, drawing the eight circles of his hell." Hundreds of small wax figures, pinned to an armature, rearranged, melted down, remade. Within two years the architectural panels dissolved. Dante gave way to Baudelaire. The figures broke their frames.

Detail: The Three Shades and The Thinker on the lintel
The Three Shades lean inward above The Poet — later renamed The Thinker. The Kiss was extracted because its tenderness was unbearable among so much suffering.Wikimedia Commons, CC BY-SA 3.0

He missed 1885. He missed 1889. He returned the money in 1895. The museum was never built. The government put a train station on the site instead — the Gare d'Orsay, opened 1900. Rodin died in Meudon on November 17, 1917, with the doors still in plaster. The Rudier foundry cast the first bronzes in 1926–28. Eight casts exist today.

The Gare d'Orsay was decommissioned in 1939, sat empty for decades, and in 1986 became the Musée d'Orsay. Inside it stands Rodin's original plaster — displayed in the museum that was built on the spot where the museum that commissioned the doors was never built. The doors arrived a hundred years late, through a different door.

The Gates of Hell from an angle, showing extreme relief depth
From the side: some figures project nearly a meter. This is sculpture pretending to be a door.Wikimedia Commons, CC BY-SA 3.0
II. Every ridge is a lens

An optical analogy—with a material boundary.

Jill's source archive models aspects of gemstone optics. Bronze is opaque and has materially different optical behavior, so gemstone theorems do not automatically transfer to it. The comparison below is a simplified visualization, not a validated bronze-optics result.

Diamond is transparent, while bronze is opaque and requires complex-index material optics. The page uses n = 0.9 + 2.6i as an illustrative input without a wavelength-specific source or uncertainty model.

17.2%
Diamond
low surface reflection, high internal transport
3.8×
65.3%
Bronze
high surface reflection, zero internal transport

Within the page's simplified calculation, the displayed bronze input yields 3.8 times the displayed diamond reflectance. This is not a measured, wavelength-resolved comparison. The qualitative visualization treats the relief as differently oriented reflectors: the sculpture becomes the illustrative optic.

Jill/Optics/Fresnel.lean:74
theorem fresnel_energy_conservation_s
    (n₁ n₂ θ₁ θ₂ : ℝ) (hn₁ : 0 < n₁) (hn₂ : 0 < n₂)
    (hcos₁ : 0 < cos θ₁) (hcos₂ : 0 < cos θ₂) :
    reflectance_s n₁ n₂ θ₁ θ₂
      + transmittance_s n₁ n₂ θ₁ θ₂ = 1
This source-model identity assumes positive real refractive indices and positive cosines. The page's complex-index bronze description is outside those displayed hypotheses, so the lemma does not certify the bronze percentages or every ridge of the sculpture.
III. The surface

Drag the light.

The photograph becomes a heightmap. The Fresnel model from jill re-lights it as bronze.

IV. The calculation

Nine hundred kilometers.

A 5-axis CNC mill carves bronze along a raster path — back and forth, like mowing a lawn. The stepover between passes sets the resolution. Smaller stepover, finer surface. Longer lawn.

The Gates have detail at every scale. Torsos are half a meter across. Ugolino's knuckle-wrinkles are a few millimeters. You need a cascade of tools, large to small, and each step down multiplies the path.

passtoolstepoverpath lengthtime
Roughing20 mm flat10.0 mm7.6 km159 hrs
Large forms20 mm ball6.0 mm21.2 km588 hrs
Faces and hands6 mm ball1.2 mm120 km5,000 hrs
Fingertips and wrinkles1 mm ball0.15 mm960 km80,000 hrs
Skin texture0.5 mm ball0.05 mm900 km150,000 hrs
Total, one machine, 24/7235,747 hrs
Under the displayed area, stepover, pass-count, and feed assumptions, the skin-texture row is a 900-kilometer raster path and about 150,000 machine-hours.

One machine, running day and night: 26.9 years.

2,456five-axis CNC mills, running in parallel, to finish by Tuesday. At $500,000 each, $1.23 billion in equipment — to reproduce a single door that a French sculptor carved with his hands.

The reason is not that the machine is slow. The reason is that Rodin's surface has detail at every scale, and the contact mechanics of a spinning tool on a metal surface are governed by a law that punishes you for going small.

V. The theorem

Why fine detail costs more than you think.

The archive contains a Hertz-contact source model. The following equality is about that definition under its stated scaling assumptions; it does not establish that a particular CNC process follows the model or derive the page's toolpath and time estimates:

Audrey/Universality/ContactMechanics.lean:82
theorem hertz_scale_invariance
    (h : HertzPair) (F lam : ℝ)
    (hF : 0 ≤ F) (hlam : 0 < lam) :
    hertzContactRadius (h.scale lam hlam) (lam^3 * F)
      = lam ^ ((4 : ℝ) / 3)
        * hertzContactRadius h F
Within the source definition, simultaneously scaling the pair and force as shown yields the displayed λ4/3 equality. Turning that contact-radius relation into a 136.8× path multiplier or a 150,000-hour estimate needs additional machining assumptions that are not proved here.

The source uses one definition for several illustrative Hertz pairs. A definitional equality does not by itself validate Hertz-model applicability, material parameters, or measurement accuracy in CNC, curling, or AFM domains:

Audrey/Universality/ContactMechanics.lean:174
theorem hertz_universality_witness :
    (∀ h : HertzPair, ∀ F : ℝ,
      hertzContactRadius h F
        = ((F * h.Rstar) / h.Estar)
          ^ ((1 : ℝ) / 3)) := by
  intro h F; rfl
The proof is rfl because the right side is the source definition. That establishes definitional identity inside the model, not empirical universality across CNC, curling, and AFM systems.
VI. The sensible methods

Three process sketches.

The following workflows, durations, staffing levels, and costs are scenario assumptions—not vendor quotes or a validated production plan. A foundry and conservation specialist would need to assess feasibility, licensing, safety, fidelity, and schedule.

Lost-wax casting — 3,000 BCE to present

Sculpt in clay. Mold in rubber. Pour wax. Coat in ceramic slurry. Burn out at 730°C. Pour bronze. Repeat 300 times. Weld. Chase. Patina. This is exactly how Rudier did it in 1926. Eight casting bays, 1.3 years. $800K–$1.2M.

3D scan + resin print + investment cast — the 2020s hybrid

Photogrammetry an existing cast. Section the mesh into 300 pieces. Print each in castable wax resin (Formlabs 3L, 50-micron layers). Invest, burn out, pour. No sculptors. The scan is the mold master.

stepequipmentpiecesparalleltotal time
3D scanArtec Leo / FARO112 days
Mesh cleanupGeomagic Wrap115 days
SLA printFormlabs 3L~30010 printers18 days
Investment castCeramic shell + kiln~3008 bays112 days
Weld + chaseTIG + grinders~300 seams4 artisans60 days
PatinaChemicals + torch115 days
Total~200 days
The bottleneck is the casting bays, not the printers. Ten Formlabs machines print all 300 sections in 18 days. The bronze pour takes four months.

About 7 months. Cost: $400K – $600K. Still not Tuesday.

Robot arm + lost-foam — the fast path

KUKA KR 210 robot arm roughs each figure in polystyrene foam in under an hour. Coat in refractory, pour bronze — the foam vaporizes (lost-foam, Shroyer 1958). Four robots, eight bays: 90 days. $500K–$700K. Still 86 days late.

VII. The business plan

Doors-R-Us: for drinking establishments in need of greatness.

You cannot deliver the Gates of Hell by Tuesday. But you can build a foundry that delivers monumental bronze doors on a repeating schedule. Drag the sliders.

The dashboard computes an illustrative minimum of user-selected capacity assumptions. Its costs, durations, demand, and revenue are not sourced forecasts. The cited source lemmas do not certify this business model.
VIII. The answer

What you tell the bar.

Seven months with the hybrid. Ninety days with robots. A year and a half the way Rudier did it. Not Tuesday. Rodin could not do it by Tuesday. He could not do it in thirty-seven years.

Under this page's chosen stepover, area, pass-count, and feed-rate assumptions, fine-detail passes dominate the estimate. The 136.8 multiplier and 900-kilometer path are scenario outputs; they are not consequences established by the cited Hertz equality or a kernel-checked fabrication model.

You could suggest a simpler door. Fewer than 180 figures. No full scope of human suffering. Something a pool cue cannot damage.

They will say no. They always say no. They saw the photograph and they want what Rodin made, which is to say they want what Dante described, which is to say they want a door that says Lasciate ogne speranza, voi ch'intrate.

Which, for a bar, is honestly perfect.

source-model boundary
This legacy page has no manifest-bound theorem receipt.

The displayed Fresnel lemma uses positive real refractive indices;
the bronze narrative invokes a complex refractive index outside that scope.

The displayed Hertz lemmas concern equalities in a source definition;
they do not derive the CNC path, duration, cost, or business forecast.

All numerical outputs remain illustrative scenario calculations.

Correction: an earlier version inferred a 136.8× CNC path multiplier directly from the Hertz scaling lemma. That inference is unsupported: the lemma simultaneously scales the modeled pair and force by λ3, and it says nothing about stepover or toolpath length. The table now stands only as an explicit scenario calculation and must not be attributed to that theorem.