Put a billiard ball, a brushed-steel spoon and a sheet of paper on the same table under the same lamp. The ball shows a tight, bright highlight; the spoon smears that highlight into a long streak; the paper has no highlight at all, just an even glow. Nothing about the light changed — only the surface changed how it redistributes the light it receives. This page is about the one function that captures that difference for any material: the BRDF, and the physically-based recipe modern renderers build on top of it.

You have already met ad-hoc shading — the diffuse + specular Phong lighting that lit early real-time graphics by hand-tuning knobs until it "looked OK". Here we do it physically: we insist the surface obeys the same conservation laws real materials obey, and those constraints are exactly what make a material look convincing under any lighting rather than only the one scene it was tuned in. The BRDF is also the material half of the rendering equation.

What the BRDF actually is

Stand at a point on a surface. Light arrives from some incoming direction \omega_i, and a camera looks back along some outgoing direction \omega_o. The Bidirectional Reflectance Distribution Function answers one question: of the light landing from \omega_i, how much heads out toward \omega_o? Precisely, it is the ratio of the reflected radiance leaving in \omega_o to the irradiance arriving from \omega_i:

That \cos\theta_i (with \theta_i the angle between \omega_i and the surface normal) is Lambert's cosine: light hitting at a glancing angle spreads its energy over more surface, so it delivers less irradiance per unit area. The BRDF is a function of four angles (two directions, each a hemisphere coordinate), and its units are inverse steradians. Every material — chalk, gold, car paint, skin — is just a different shape of f_r.

The two extremes: diffuse and specular

The simplest legal BRDF is the Lambertian (perfectly diffuse) surface: it scatters incoming light equally in all outgoing directions, so f_r is a constant that does not depend on \omega_i or \omega_o at all. If the surface reflects a fraction \rho of the light it receives (its albedo, with 0 \le \rho \le 1), energy conservation pins the constant exactly:

Where does the \pi come from? Integrate a constant c against \cos\theta_o over the hemisphere and you get c\pi; forcing that to equal \rho gives c = \rho/\pi. Beginners who forget the \pi get a surface that is too bright by a factor of about three — a classic energy leak.

The opposite extreme is a perfect mirror (specular): light from \omega_i leaves in exactly one direction, the mirror reflection, and nowhere else. Its BRDF is a spike — a delta function — infinitely tall and infinitely thin. Real materials live between these two: a glossy surface has a diffuse base plus a specular lobe that is a blurred, finite-width version of the mirror spike. The width of that lobe is the whole game, and physically-based rendering earns the blur from first principles.

Microfacets: a rough surface is a sea of tiny mirrors

Physically Based Rendering (PBR) explains gloss with one idea: zoom in far enough and a "rough" surface is not smooth at all — it is a landscape of millions of microscopic facets, each a perfect mirror pointing in a slightly different direction. You see a broad soft highlight not because any facet blurs light, but because the facets' normals are scattered, so different facets flash the light toward your eye from a whole range of surface points. A polished surface has facets nearly aligned with the macroscopic normal (a tight highlight); a rough surface has wildly varying facet normals (a wide, soft sheen).

The Cook–Torrance microfacet BRDF turns that picture into three factors. Only facets whose normal points exactly along the half-vector h = \tfrac{\omega_i + \omega_o}{\lvert \omega_i + \omega_o \rvert} can mirror light from \omega_i to \omega_o. So the specular BRDF is:

The 4\cos\theta_i\cos\theta_o denominator is the geometric bookkeeping that converts between facet-space and surface-space solid angles. Crucially, this BRDF is built to be reciprocal and energy-conserving by construction — the physics is baked in, not tuned in.

Watch the lobe widen

Below is a cross-section of the GGX distribution D as a function of the angle between a facet's normal and the surface normal. Drag the roughness slider. At low roughness the curve is a tall, narrow spike — nearly all facets aligned, so the highlight is a crisp mirror dot. As roughness climbs the peak collapses and the curve flattens into a broad hump: the same total energy, now smeared across a wide range of angles — the sharp highlight melts into a soft satin sheen. That single knob, from \alpha \to 0 (chrome) to \alpha \to 1 (chalky), sweeps a material through its entire range of gloss.

Notice the area under the curve stays roughly fixed as it spreads — that is energy conservation visible as a picture. A rough surface does not reflect less specular light overall; it just distributes the same light over a wider cone, so any one direction gets a dimmer share.

Worked example: roughness, and the grazing-angle trick

Widening the lobe. Take a GGX surface and push \alpha from 0.05 to 0.6. At \alpha = 0.05 the distribution's peak is enormous — the highlight is a pinpoint reflection of the light source, a chrome bearing. At \alpha = 0.6 the peak value drops by more than an order of magnitude and the lobe fans out over tens of degrees: the light source is no longer recognisable, just a soft bright smear the size of a fist. The transition is smooth and monotone — no re-tuning, no per-light hacks — which is exactly why one roughness map can describe a scuffed metal panel that reads correctly under a studio softbox and a hard noon sun.

Fresnel makes even a rough surface a mirror at grazing angles. Take a rough dielectric — unpolished plastic, dry asphalt — with a tiny head-on reflectance F_0 \approx 0.04 (about 4%, typical of non-metals). Look straight down at it (\theta = 0): Schlick gives F = 0.04, barely reflective, and you see its dull diffuse colour. Now slide your eye to a grazing angle, \theta = 80^\circ, so \cos\theta \approx 0.174:

Reflectance has leapt from 4% to about 41% — an order of magnitude — purely from viewing angle. This is why a dull road turns into a blinding mirror in low evening sun, why the far edge of a matte table gleams, and why a lake is glassy near the far shore. Every surface, however rough, becomes mirror-like at a shallow enough angle. Fresnel is the term that captures it, and its absence is the single most common giveaway of a "fake"-looking render.

The metallic / roughness workflow

PBR's constraints did something quietly revolutionary for artists: they made materials portable. Because a physically-based BRDF behaves correctly under any lighting, studios converged on a tiny, intuitive set of authored channels — the metallic/roughness workflow:

Three or four sliders, and the same material dropped into a night street, a sunset, or a spaceship interior looks like the same stuff every time. Before PBR, a material was tuned to a scene and broke when the lighting changed; after PBR, a material is a description of a surface, and the renderer figures out the rest. That consistency — energy conservation plus Fresnel plus a single roughness knob — is what makes PBR read as "real".