Multiscattering Energy Compensation in GGX GLSL

Standard microfacet bidirectional reflectance distribution function (BRDF) models such as GGX assume that incident light bounces off a single microfacet before either escaping the surface or becoming occluded. While this single-scattering approximation works well for smooth surfaces, it fails dramatically at high roughness values where multiple internal reflections occur. By neglecting these secondary bounces, the model discards energy, leading to unnatural darkening and saturation loss at high roughness. Multiscattering energy compensation solves this problem by deriving and reintroducing the missing scattered energy back into the specular lobe, ensuring energy conservation across all roughness ranges in GLSL shaders.

The Mechanism of Energy Loss in Single-Scattering GGX

In a standard Cook-Torrance GGX formulation, the specular term depends on three components: the normal distribution function (\(D\)), the Fresnel reflectance (\(F\)), and the geometric masking-shadowing function (\(G\)). The \(G\) term accounts for microfacets blocking incoming light (shadowing) or reflected light (masking).

When light is shadowed or masked, standard GGX treats that radiant energy as absorbed or completely lost. In reality, light reflected into a neighboring microfacet does not disappear; it undergoes multiple subsequent bounces until it eventually exits the micro-surface cavity. At high roughness values (where microfacet valleys are steep and deep), the probability of multiple bounces approaches 100%. Without modeling these inter-reflections, the total directional albedo of the surface plummets far below theoretical limits, making rough metals and rough dielectrics appear abnormally dark and washed out, especially near grazing angles.

Deriving the Multiscattering Term

To compensate for this missing energy without incurring the heavy computational cost of stochastic path tracing per pixel, real-time rendering commonly employs the Kulla-Conty approximation.

  1. Directional Albedo (\(E_\mu\)): First, the directional albedo of the single-scattering GGX lobe is calculated for an incoming direction with cosine \(\mu = N \cdot V\): \[E(\mu) = \int_{\Omega} f_{ss}(L, V) (N \cdot L) \, dL\] This represents the fraction of energy successfully reflected in a single bounce. The missing energy from the first bounce is therefore \(1 - E(\mu_o)\) for the outgoing direction and \(1 - E(\mu_i)\) for the incident direction.

  2. Average Albedo (\(E_{avg}\)): Integrating \(E(\mu)\) over the hemisphere yields the average albedo: \[E_{avg} = 2 \int_0^1 E(\mu) \mu \, d\mu\]

  3. Multiscattering Addition (\(f_{ms}\)): By assuming infinite isotropic bounces within the micro-geometry, the multiple-scattering lobe \(f_{ms}\) can be expressed as: \[f_{ms} = \frac{(1 - E(\mu_o))(1 - E(\mu_i))}{\pi (1 - E_{avg})}\]

  4. Fresnel Handling for Conductors and Dielectrics: Because each subsequent bounce interacts with the surface material, it absorbs energy according to the material's Fresnel reflectance. The average Fresnel response is scaled by an average Fresnel coefficient: \[F_{avg} = 2 \int_0^1 F(\mu) \mu \, d\mu\] The multi-scattered light is modulated by a geometric series incorporating \(F_{avg}\) to correctly preserve the color of rough metals and the white highlights of rough dielectrics.

Implementation in GLSL

In real-time GLSL implementations, calculating the integrals on the fly is prohibitive. Instead, engines precompute \(E(\mu)\) and pack it into a 2D Lookup Table (LUT), usually indexed by roughness and the dot product of the normal and view vector (\(N \cdot V\)).

Below is a standard GLSL implementation combining single-scattering GGX with Kulla-Conty multiscattering energy compensation:

// Precomputed BRDF LUT provides:
// Red channel: Directional albedo scale (or E(NdotV))
// Green channel: Directional albedo bias
// Blue channel: Average directional albedo E_avg (optional, or computed analytically)
uniform sampler2D u_BRDFLUT;

vec3 F_Schlick(float cosTheta, vec3 F0) {
    return F0 + (1.0 - F0) * pow(clamp(1.0 - cosTheta, 0.0, 1.0), 5.0);
}

vec3 F_Avg(vec3 F0) {
    return F0 + (vec3(1.0) - F0) * (1.0 / 21.0);
}

vec3 ComputeMultiscatteringGGX(
    vec3 N, vec3 V, vec3 L, 
    float roughness, vec3 F0, 
    vec3 singleScatterBRDF) 
{
    float NdotV = clamp(dot(N, V), 0.001, 1.0);
    float NdotL = clamp(dot(N, L), 0.001, 1.0);

    // Sample single-bounce directional albedo from the LUT
    vec2 brdfSampleV = texture(u_BRDFLUT, vec2(NdotV, roughness)).rg;
    vec2 brdfSampleL = texture(u_BRDFLUT, vec2(NdotL, roughness)).rg;

    // Directional albedos for view and light directions
    float E_V = brdfSampleV.r + brdfSampleV.g;
    float E_L = brdfSampleL.r + brdfSampleL.g;

    // Approximate hemisphere-averaged albedo
    float E_avg = texture(u_BRDFLUT, vec2(1.0, roughness)).r; 

    // Missing energy factor
    float factor = max(1.0 - E_avg, 1e-5);
    vec3 f_ms = vec3((1.0 - E_V) * (1.0 - E_L)) / (3.14159265 * factor);

    // Fresnel contribution over infinite bounces
    vec3 F_avg_val = F_Avg(F0);
    vec3 F_ms = (F_avg_val * F_avg_val * E_avg) / (vec3(1.0) - F_avg_val * (1.0 - E_avg));

    // Combine single scattering with compensated multiple scattering
    vec3 multiScatterBRDF = f_ms * F_ms;

    return singleScatterBRDF + multiScatterBRDF;
}

Visual and Physical Impact

Integrating multiscattering compensation ensures that as surface roughness increases toward 1.0, the total reflected irradiance approaches the theoretical conservation limit dictated by the Fresnel equation rather than decaying toward zero. The visual result is immediately noticeable: rough gold, copper, and rough plastics maintain their proper luminance and edge vibrancy, completely removing the artificial "dark ring" artifacts typical of conventional Cook-Torrance GGX shaders.