Fix furnace test energy loss for intermediate metalness values

[mrdoob]

Description

Fixes energy conservation in the white furnace test when metalness values are between 0 and 1. Previously, intermediate metalness values (especially 0.5-1.0 roughness) were losing energy due to incorrect application of the multiscattering compensation formula.

Problem

The Fdez-Agüera multiscattering formula uses Favg = F0 + (1 - F0) * (1/21), which is designed for pure dielectric OR conductor F0 values, not blended ones. The previous implementation applied this formula to specularColor, which was already blended between dielectric (0.04 or IOR-based) and metallic (base color) F0 values based on metalness. This caused incorrect energy compensation for intermediate metalness values.

Solution

Following the approach used in EEVEE and Frostbite:

1. Compute multiscattering separately for dielectric and metallic paths:
   • Dielectric path: F0 = 0.04 (or IOR-based for physical materials)
   • Metallic path: F0 = base color
2. Mix the single-scattering and multi-scattering results based on metalness
3. Use the dielectric path for diffuse energy conservation

Changes

Material Structure (PhysicalMaterial)

Reorganized and renamed fields for clarity:

• diffuseColor - Original base color (before metalness reduction)
• diffuseContribution - Diffuse component reduced by metalness: baseColor * (1 - metalness)
• specularColor - Dielectric F0 (0.04 or IOR-based)
• specularColorBlended - Blended F0: mix(specularColor, diffuseColor, metalness) for direct lighting
• metalness - Metalness factor

Indirect Specular Lighting (RE_IndirectSpecular_Physical)

• Compute separate dielectric and metallic multiscattering contributions
• Mix results based on metalness:
  • vec3 singleScattering = mix(singleScatteringDielectric, singleScatteringMetallic, metalness);
  • vec3 multiScattering = mix(multiScatteringDielectric, multiScatteringMetallic, metalness);
• Use dielectric total scattering for diffuse energy conservation

Direct Specular Lighting (BRDF_GGX_Multiscatter)

• Updated to use specularColorBlended for the blended specular reflection

Other Updates

• All diffuse lighting now uses diffuseContribution instead of the old diffuseColor
• Iridescence evaluation uses specularColorBlended
• Transmission uses diffuseContribution and specularColorBlended

Testing

The white furnace test (examples/webgl_furnace_test.html) now correctly preserves energy across all metalness (0-1) and roughness (0-1) combinations.

Known Issues

• The metallic multiscattering path currently uses computeMultiscatteringIridescence when iridescence is enabled, which is physically incorrect (iridescence should only affect dielectric reflections, not metallic ones). This will be addressed in a future PR.

References

• Fdez-Agüera's "Multiple-Scattering Microfacet Model for Real-Time Image Based Lighting"
• EEVEE implementation approach

Scope

This PR addresses the WebGL renderer only. WebGPU renderer fixes will be addressed in a separate PR.

Before	After

(Made using Claude Code Web)

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	355.14 86.33	356.38 86.5	+1.23 kB +174 B
WebGPU	609.66 171.01	609.66 171.01	+0 B +0 B
WebGPU Nodes	608.27 170.75	608.27 170.75	+0 B +0 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

	Before	After	Diff
WebGL	486.77 121.07	488 121.24	+1.23 kB +167 B
WebGPU	678.52 186.29	678.52 186.29	+0 B +0 B
WebGPU Nodes	620.51 169.56	620.51 169.56	+0 B +0 B

[mrdoob]

@sunag

Couldn't get Claude to port the fix to the WebGPURenderer. Will try in another PR.

[mrdoob]

Claude spotted an issue with the iridescence code while working on this:

computeMultiscatteringIridescence( geometryNormal, geometryViewDir, material.diffuseColor, ..., material.iridescence, material.iridescenceFresnel, ... )

This is physically incorrect. Iridescence should only affect dielectric specular reflections, not metallic reflections. When you have a pure metal (metalness = 1), iridescence shouldn't affect it at all.

The metallic path should always use the non-iridescence version because metals don't exhibit thin-film interference effects - they have high absorption and the thin-film model doesn't apply.

I'll leave this for another PR.

/fyi @WestLangley