Why Surface Roughness Matters in Optics
Every real surface has microscopic height variations. When light interacts with these imperfections, some fraction of the incident power is redirected away from the ideal specular (mirror-like) direction. This redirected light is scatter, and it is the dominant source of stray light, veiling glare, ghost images, and contrast loss in optical and electro-optical systems. Understanding and controlling scatter is central to optical and opto-mechanical engineering — from telescope mirrors and lithography optics to laser cavities and precision instrument housings.
The key insight is that a single roughness number (Ra or Rq) is not enough to predict how a surface will scatter light. Two surfaces with identical reported Ra can produce dramatically different scatter patterns depending on the spatial-frequency content of the roughness, the wavelength of light, the angle of incidence, and — critically — the bandwidth of the metrology instrument that measured the roughness in the first place.
Surface Roughness Parameters
Ra (arithmetic average roughness) is the mean absolute height deviation from the surface mean line. It is the most commonly specified roughness parameter in manufacturing drawings and procurement specs, but it discards all information about the spatial structure of the roughness. Two surfaces — one with gentle long-wavelength undulations and one with sharp short-wavelength texture — can have the same Ra yet behave completely differently optically.
Rq (RMS roughness) is the root-mean-square height deviation. It weights larger deviations more heavily than Ra and is the parameter that appears directly in optical scatter theory. For Gaussian-distributed surfaces, Rq ≈ 1.11 × Ra. When specifying optics, Rq is generally preferred over Ra because of this direct connection to scatter physics.
Rz (peak-to-valley) captures the maximum excursion over the evaluation length. It is sensitive to isolated defects (scratches, pits, tool marks) that Ra and Rq may average away, making it a useful complement for quality control and for flagging surfaces whose roughness statistics are dominated by rare outlier features.
Practical note for mechanical engineers: Manufacturing process specifications (grinding, lapping, polishing, bead-blasting, anodizing) often call out Ra on drawings per ASME B46.1 or ISO 4287. These values are meaningful for fit, wear, and sealing purposes but do not directly tell an optical engineer what the surface will do to a beam. Bridging this gap is a central purpose of this tool.
Total Integrated Scatter (TIS)
TIS is the fraction of reflected (or transmitted) power that is scattered out of the specular beam. For a smooth surface with Gaussian roughness statistics illuminated at wavelength λ and angle of incidence θ, the Rayleigh–Rice perturbation theory gives:
Smooth-surface TIS (Rayleigh–Rice)
This formula is accurate when the phase perturbation is small — specifically, when 4πσ cos θ / λ ≪ 1 (the "smooth-surface" regime). As roughness increases relative to wavelength, the approximation breaks down and TIS saturates toward unity, meaning essentially all light is scattered and the surface becomes a diffuse reflector.
Crucially, σ in this equation is the bandlimited RMS roughness — the roughness evaluated only over the spatial-frequency range that actually contributes to scatter at the given wavelength. This is where metrology bandwidth enters the picture.
What TIS does and does not tell you: TIS is a scalar — it reports the total fraction of power lost from specular, but says nothing about where that scattered light goes. A surface with TIS = 0.5% that scatters into a narrow ring one degree from specular has completely different stray-light consequences than one with TIS = 0.5% scattered uniformly over the hemisphere. For system-level analysis, the angular distribution (BRDF) matters at least as much as the total.
The Power Spectral Density (PSD) and Why It Matters
The power spectral density (PSD) decomposes surface roughness into contributions at each spatial frequency. Low spatial frequencies correspond to long-wavelength figure errors (form); mid spatial frequencies produce small-angle scatter near specular; high spatial frequencies create wide-angle scatter. The PSD is the Fourier transform of the surface autocovariance function, and it is the single most complete description of a surface's scattering potential.
The connection between PSD and scatter is direct: in the Rayleigh–Rice regime, the BRDF at any scatter angle is proportional to the PSD evaluated at the corresponding spatial frequency. The total scatter (TIS) is the integral of the PSD over the relevant frequency band, weighted by the optical transfer function:
PSD–scatter integral (left) and PSD–BRDF relationship (right), 1-D, normal incidence
This means a surface with most of its roughness at low spatial frequencies (long correlation length) will scatter primarily into a narrow cone near specular, while a surface with significant high-frequency content (short correlation length) will scatter more broadly. Two surfaces with the same RMS roughness but different PSD shapes will produce very different angular scatter distributions — a fact that Ra/Rq alone cannot capture.
In practice, very few surface specifications include PSD data — but for critical stray-light and scatter-sensitive systems, it is the right thing to ask a vendor for. A PSD measurement (typically from a white-light interferometer or a stitched set of measurements at different magnifications) gives you far more design confidence than an Ra number from an uncontrolled stylus trace.
BRDF and BSDF
The bidirectional reflectance distribution function (BRDF) quantifies how much light is scattered into each direction per unit solid angle per unit incident irradiance:
BRDF definition (sr⁻¹)
For transmissive surfaces, the equivalent quantity is the BTDF (bidirectional transmittance distribution function). Together they form the BSDF (bidirectional scatter distribution function).
BRDF is the gold standard for specifying scatter performance because it captures the full angular distribution, not just the scalar total. A surface with TIS = 1% concentrated in a narrow ring near specular is very different from one with TIS = 1% spread over the full hemisphere — even though TIS alone cannot distinguish them.
When to request BRDF data: For any surface that sits in the optical path and has TIS above ~0.1%, or for any baffle or housing surface where wide-angle scatter rejection matters, BRDF measurement data (often at 633 nm from a scatterometer like a Schmitt Measurement Systems or Surface Optics instrument) gives you far more information than a roughness spec. In stray-light analysis codes (FRED, TracePro, Zemax), surfaces are modeled by BRDF — not by Ra.
Measurement Bandwidth — The Hidden Variable
Every surface roughness measurement instrument has a finite spatial-frequency bandwidth, bounded by the scan length at the low end and by the tip radius or pixel size at the high end. A stylus profilometer with a 5 mm scan length and a 1 µm tip radius captures spatial frequencies from roughly 0.2 to 500 mm⁻¹. An AFM scanning a 50 µm field captures 20 to 50,000 mm⁻¹. A white-light interferometer falls somewhere in between, and can be configured with different objectives and fields of view.
Because the reported Rq is computed from the measured profile, it reflects only the roughness within the instrument's bandwidth. If the instrument cannot see the spatial frequencies that matter for scatter at your operating wavelength, the reported Rq will be misleading — potentially too low (missing high-frequency scatter drivers) or too high (dominated by long-wavelength form error that doesn't produce scatter).
This is why a supplier's roughness certificate must always be interpreted in the context of the measurement conditions. The same polished optic measured by a stylus profilometer and a white-light interferometer will often yield different Rq values, not because the surface changed, but because the instruments sample different frequency bands. A specification that reads "Ra < 2 nm" without stating the instrument type, scan length, and cutoff filter is essentially ambiguous.
Correlation Length and PSD Shape
The correlation length is the lateral distance over which surface heights remain statistically correlated. A long correlation length means the roughness is dominated by gentle, slowly-varying undulations; a short correlation length means the surface has fine-scale texture. In PSD terms, the correlation length controls the "knee" frequency where the PSD transitions from a flat plateau to a power-law roll-off.
Together, correlation length and PSD slope define the angular "shape" of scatter. This tool models the PSD as a modified Lorentzian (ABC model): flat below the correlation frequency, then falling as fn above it — a reasonable approximation for many polished and machined surfaces, though real-world PSDs can be far more complex (e.g., diamond-turned surfaces with discrete periodic components, or ground surfaces with bimodal frequency distributions).
PSD Slope — Deep Dive
The PSD slope (the exponent n of the power-law tail, typically in the range −1 to −4) is one of the most physically consequential surface parameters — yet it almost never appears on a drawing or vendor data sheet. It controls how rapidly roughness power falls off at high spatial frequencies, and it has a direct, quantitative effect on both the magnitude and angular shape of optical scatter.
What the slope number means physically: On a log-log PSD plot, the slope is the gradient of the straight-line portion at frequencies above the correlation-length knee. A slope of −1 means roughness power falls very slowly with frequency — the surface has nearly equal energy at every scale, producing broad, uniform scatter across all angles. A slope of −4 means high-frequency power drops off extremely fast — almost all roughness is at long spatial periods, and scatter is concentrated in a tight cone near specular. The intermediate values (−2 to −3) produce the moderate fall-off typical of most real polished and machined surfaces.
Why it matters for system design: Consider two surfaces, both with Rq = 2 nm and correlation length = 10 µm, but one with PSD slope −1.5 and the other with slope −3.0. At λ = 633 nm they will report nearly identical TIS (because TIS integrates over the entire PSD and the total area under both curves is the same, being set by Rq). However, the −1.5 surface distributes its scatter broadly — it produces a measurable BRDF floor out to large angles, causing veiling glare and detector haze. The −3.0 surface concentrates its scatter close to specular — the far-field BRDF drops off rapidly, but the near-specular "skirt" is much brighter. In an imaging system, the −1.5 surface degrades large-area contrast, while the −3.0 surface degrades fine-detail contrast (MTF at mid-frequencies). Same Rq, same TIS, completely different system impact.
PSD Slopes by Surface Process
Different manufacturing processes produce characteristic PSD slope ranges. These are approximate and depend on specific tooling, materials, and process parameters, but they are useful as engineering starting points:
Superpolished optics (slope −2.5 to −3.5): Ion-beam-figured or magneto-rheological-finished (MRF) optics achieve very steep PSD slopes because the finishing process preferentially removes mid- and high-frequency roughness. The resulting scatter is tightly concentrated near specular. These surfaces are used in gravitational-wave detector mirrors (LIGO), ring-laser gyroscopes, and high-finesse Fabry–Pérot cavities where wide-angle scatter must be suppressed below 10⁻⁶ sr⁻¹. A superpolished mirror might report Rq = 0.1 nm with slope −3.0, producing TIS on the order of 10⁻⁵ at 633 nm.
Conventional pitch-polished optics (slope −2.0 to −2.5): Standard optical polishing with pitch laps produces a moderate PSD slope. These surfaces are the workhorses of precision optics — camera lenses, telescope mirrors, laser windows. The PSD typically shows a clean power-law from ~1 mm⁻¹ to ~1000 mm⁻¹ with slope near −2. At 633 nm with Rq = 1 nm, TIS is around 0.04% — entirely adequate for most imaging and laser applications.
Diamond-turned surfaces (slope −2.0 to −2.5, plus periodic spikes): Single-point diamond turning produces a surface whose PSD has a smooth background (slope typically −2 to −2.5) overlaid with sharp peaks at the tool-mark spatial frequency and its harmonics. The periodic component does not follow a power-law at all — it produces discrete diffraction orders (ghosts) at angles determined by the tool-mark pitch. A 10 µm tool-mark pitch at λ = 1550 nm produces first-order diffraction at ±9°. The PSD slope of the background matters for the continuous scatter floor, but the periodic peaks dominate the far-field pattern. This is why specifying only Ra/Rq is particularly misleading for diamond-turned surfaces.
Ground and lapped surfaces (slope −1.5 to −2.0): Mechanical grinding and lapping leave surfaces with relatively shallow PSD slopes. The abrasive process creates roughness at many spatial scales simultaneously, so power does not fall off quickly at high frequencies. This produces broad-angle scatter. Ground glass diffusers exploit this intentionally — a ground surface at λ = 550 nm with Rq = 500 nm and slope −1.8 scatters light into a wide cone, functioning as an effective diffuser. For a precision optic, a residual grinding texture with slope −1.5 is a problem: even if subsequent polishing reduces Rq substantially, the shallow slope means the remaining roughness is spectrally "flat" and produces scatter at all angles.
Bead-blasted and textured metals (slope −1.0 to −1.5): Bead blasting, chemical etching, and sand casting produce surfaces with very shallow PSD slopes — roughness power is nearly uniform across all measured frequencies. The resulting BRDF is approximately Lambertian (uniform in angle), which is actually desirable for stray-light baffles and optical blackening. A bead-blasted aluminum housing wall with Rq = 1000 nm and slope −1.2 scatters almost all incident light diffusely, with no measurable specular component. The very shallowness of the slope is what makes these surfaces effective absorbers in baffle systems.
Anodized and black-coated surfaces (slope −1.0 to −1.8): Black anodize, Aeroglaze Z306, and similar coatings are designed to maximize absorption and wide-angle diffuse scatter. Their PSDs tend to be shallow-sloped with high Rq values. For these surfaces, the BRDF at large angles (60–85° from normal) is more important than TIS — a good baffle coating needs low BRDF at grazing angles to prevent "specular leak" past the baffle edge. PSD slope affects this: a very shallow slope (−1.0) tends to produce more uniform angular scatter including at grazing, while steeper slopes can leave a residual near-specular lobe that compromises baffle performance.
Contaminated surfaces (variable slope, often −1.5 to −2.5): Particulate contamination adds a broadband PSD component whose slope depends on the particle size distribution. Fine particulates (sub-micron) add high-frequency PSD content and steepen the apparent slope; coarse particulates add low-frequency bumps and flatten it. Molecular contamination (fingerprints, outgassing films) tends to add a smooth, broad PSD hump at mid-frequencies. In either case, the contamination contribution is additive to the substrate PSD and can dominate scatter even on a pristine substrate.
Reading a PSD Slope from Measurement Data
If you have PSD data from a vendor or your own metrology lab, extracting the slope is straightforward: plot log(PSD) versus log(spatial frequency) and fit a straight line to the power-law region above the correlation knee. Most white-light interferometer software (Zygo MetroPro/Mx, Bruker Vision64, 4D Technology) can export PSD data directly. A few practical cautions: the slope may not be constant across the full frequency range (many real surfaces show a "broken" PSD with different slopes in different bands); instrument noise can flatten the PSD at the highest frequencies; and windowing artifacts can distort the PSD at the lowest frequencies near the edge of the measurement aperture. For reliable slope determination, use at least one decade of spatial frequency in the fit range and confirm that the result is not dominated by instrument noise floor.
Scatter Regimes
This tool classifies surfaces into three regimes based on the Rayleigh parameter 4πσ cos θ / λ:
Smooth-surface regime (parameter ≪ 0.1): The Rayleigh–Rice perturbation theory is accurate. TIS is small, the specular beam is well-defined, and scatter is a minor perturbation. Most precision optics — mirrors, lenses, windows, prisms — live here when properly polished.
Intermediate regime (parameter 0.1 – 0.5): The perturbation model begins to overestimate scatter. Real surfaces may show non-Gaussian effects. Results should be treated as approximate and validated against scatter measurement data. Diamond-turned surfaces with moderate tool marks, lightly ground surfaces, and optics at short UV wavelengths often fall in this zone.
Rough-surface regime (parameter > 0.5): The specular beam is largely destroyed. The surface acts as a diffuse scatterer and the TIS formula is no longer meaningful — essentially all light is scattered. In this regime, the angular distribution (BRDF shape) and total hemispherical reflectance matter far more than a TIS number. Baffles, diffusers, bead-blasted housings, and anodized mechanical surfaces typically fall here.
Common Pitfalls in Practice
Specifying Ra without bandwidth: A drawing callout of "Ra < 5 nm" is common but optically incomplete. Without knowing the spatial-frequency window of the measurement, two vendors can comply with the spec using different instruments and deliver surfaces with very different scatter performance. Always pair an Ra or Rq spec with an instrument type and scan-length requirement, or (better) reference ISO 10110-8 which explicitly addresses mid-spatial-frequency roughness in an optical context.
Assuming Ra predicts scatter: Ra is an amplitude-only statistic. It does not encode spatial frequency content, periodicity, or isotropy. A surface with Ra = 3 nm from gentle figure residuals will scatter orders of magnitude less than a surface with Ra = 3 nm from fine periodic tool marks at a spatial frequency that happens to diffract light into your detector.
Ignoring mid-spatial-frequency (MSF) errors: MSF roughness — spatial periods from roughly 0.1 mm to 10 mm — is a well-known trouble zone in precision optics. It is often missed by interferometers that filter out these scales, and by profilometers that lack the lateral resolution or scan length to capture them fully. MSF error produces small-angle scatter that degrades contrast and is particularly problematic in imaging systems, coronagraphs, and high-energy-density laser optics.
Conflating roughness with cleanliness: Particulate and molecular contamination can dominate scatter, even on a perfectly polished surface. The contamination scatter contribution is typically additive in TIS and broadband in angle. A "contaminated mirror" in this tool's preset illustrates how a small increase in effective Rq from surface contamination can produce significant stray-light consequences — especially at short wavelengths.
When to Use This Tool — and When Not To
This tool is designed for first-order trade studies, design reviews, and supplier conversations. It provides defensible engineering estimates of scatter given surface roughness parameters and metrology conditions, and it highlights the regimes where simple models break down. It is appropriate for comparing surface-finish options during preliminary design, for sanity-checking vendor roughness data, for building intuition about scatter-wavelength-roughness relationships, and for communicating to mechanical-design colleagues why an optical surface spec requires more than an Ra callout.
This tool is not a substitute for rigorous scatter modeling. For critical stray-light analysis, flight-hardware qualification, or high-performance system verification, measured BRDF data and a proper stray-light analysis code (FRED, TracePro, or equivalent) are required. The simplified PSD model used here (isotropic, single-slope, Gaussian autocorrelation) does not capture the full complexity of real surfaces — particularly anisotropic textures, discrete periodic structures, or multi-scale roughness with distinct process signatures at different spatial-frequency bands.