Optical windows are classified along four specification dimensions: the wavelength range they must transmit, the substrate material, the coating on their surfaces, and the size and thickness of the part. That is the classification that matters when you have to write down a part number. Lists of window shapes exist too, but shape describes geometry, not fitness for your beam path.
The order of the dimensions does most of the work. Wavelength eliminates most substrates before cost ever enters the decision. The surviving materials differ in hardness, thermal behavior, and refractive index, and those differences set your coating options. Aperture and pressure load then fix the size and thickness. Walk the four dimensions in that order and a thousand-line catalog collapses into a short, comparable list - which is what this guide does, one dimension at a time.
Classified by specification, not by shape
An optical window is a flat, transparent plate designed to maximize transmission in a specified wavelength range while minimizing reflection and absorption [1]. It separates two environments - a sensor and the world outside it - and adds no optical power of its own [1]. Selection therefore comes down to material, surface, and mechanical properties rather than focal behavior.
Most catalogs lead with shape: flat windows, wedged windows, domes, Brewster windows. That axis is real, but it classifies geometry. It says nothing about whether the part survives your wavelength, your pressure differential, or your throughput budget.
Specification dimensions answer those questions, and there are four of them. Wavelength range comes first, because every substrate has a hard transmission cut-off. Material comes second, because the surviving candidates differ in hardness, expansion, and index. Coating comes third, since the substrate’s index decides what reflection costs you and which coating classes are worth buying. Size and thickness come last, set by your clear aperture and your pressure load. Filter in that order and each step hands a smaller, more uniform group to the next.
Dimension 1: Wavelength range
The spectral bands that sort windows
Suppliers already band their catalogs by spectrum: ultraviolet windows, visible windows, infrared windows [2]. One physical fact drives that sorting - each substrate transmits only between two cut-offs, and nothing outside the range matters.
Those cut-offs are unforgiving. UV-grade fused silica reaches below 200 nm, which is why it owns the ultraviolet class [3]. Sapphire spans about 150 nm to 4.5 µm and calcium fluoride about 130 nm to 9.5 µm [3]. Zinc selenide runs from 0.6 to 21 µm [4]. Germanium covers 2 to 16 µm while blocking the UV, the visible, and most of the near-IR outright [3]. That blindness is a feature in thermal imaging, where stray visible light would contaminate the measurement.
Your first classification question is therefore not “which window?” but “which band?”. Answer it and most of the catalog disappears before you have compared a single price.
Broadband versus single-band windows
Band width matters as much as band location. A window for one laser line needs high transmission at a single wavelength, and almost any glass or fused silica part qualifies. A window for a spectrometer covering 0.4 to 5 µm needs one substrate that spans visible and mid-infrared, which shortens the list to materials like calcium fluoride and sapphire [3].
Broadband instruments push toward spanning substrates even when a specialist would win in a single band. Zinc selenide exemplifies the trade: it owns the thermal bands, yet still leaks enough light through from about 600 nm to pass a red alignment laser [4]. One part then serves two jobs, which is exactly what a multi-spectral instrument wants.
Dimension 2: Substrate material
Three material families and what each is for
Within your band, several substrates usually still transmit, and the industry groups them into three families. RP Photonics describes the split. Glasses like fused silica and BK7 serve visible or near-infrared light. At longer wavelengths, crystalline materials such as calcium fluoride join semiconductors like zinc selenide, silicon, and germanium [2].
Each family carries a personality. Glasses are inexpensive, isotropic, and easy to polish to high quality. Crystals reach bands glasses cannot and survive environments glasses will not. IR semiconductors own the thermal bands but are soft and depend on coatings for usable throughput [3].
The tie-breaker properties
Hardness settles abrasive and pressure-loaded applications. Sapphire sits at a Knoop hardness of 2200 - behind only diamond among common window materials - while sodium chloride sits near 18 and the fluorides are softer still [3]. High Knoop hardness also tracks the ability to withstand pressure differentials [1]. Calcium fluoride illustrates the cost of softness: it spans UV to mid-IR, yet demands careful edge design, cleaning, and thermal management [5].
Thermal behavior settles the rest. Fused silica expands by roughly 0.5 x 10-6 per kelvin, several times less than typical optical glasses, so it holds its figure under thermal cycling [6]. Zinc selenide brings good thermal-shock resistance to a soft IR part [4]. One more subtlety: sapphire and crystalline quartz are anisotropic, so the crystal cut becomes part of the specification, not just the material name [5].
Dimension 3: Coating
Uncoated windows and the 4 percent tax
Every uncoated surface between air and ordinary glass reflects about 4 percent of the incident light through Fresnel reflection, and a window has two surfaces. Edmund Optics puts the result plainly: total transmission through an uncoated glass substrate is only about 92 percent [7]. If losing roughly 8 percent of your beam is acceptable, uncoated is a legitimate class - and often the cheapest one.
Refractive index changes the arithmetic. Surface loss grows with index, and it is “especially important for high-index infrared materials such as silicon and germanium” [5]. Zinc selenide illustrates the scale: its index is about 2.4 at 10.6 µm [4], versus roughly 1.5 for common glasses. Each bare ZnSe surface therefore throws away far more than 4 percent. In the IR classes the coating is not an upgrade. It is what makes the window usable.
Single-band, broadband, and functional coatings
Coated windows divide first by band coverage. Single-band designs - often V-coats - serve one laser line at near-zero reflection. Broadband designs hold low reflectance across a wide spectral range. Edmund’s standard visible 0° coating cuts average reflection to 0.4 percent over 425-675 nm [7]. Many broadband designs specify 0.5 percent average or less per surface. Choose by your spectrum, not by the best number in a data table.
A third group is functional. Dichroic and beamsplitting coatings split bands instead of just passing them. Protective finishes exist too - one supplier offers diamond-like carbon on windows alongside standard AR stacks [2]. Laser-safety viewing windows, which block the laser line while passing visible light, belong to this class as well [2].
Dimension 4: Size, thickness, and clear aperture
Diameter, thickness, and clear aperture
Dimensional classification starts prosaically. Stock windows cluster around standard round diameters, and round parts cost less than rectangles or profiles with cut-outs. The spec lines that separate a precision window from a commodity one are thickness tolerance and parallelism. Parallelism is often quantified in arcseconds, with tight parts specified below 1 arcsecond [2].
Clear aperture is the quiet trap in this dimension. It is “the diameter or size of an optical component that must meet specifications,” and outside it the manufacturer guarantees nothing [8]. Bevels and edges sit in that unguaranteed zone, so your beam must fit inside the clear aperture, not the nominal diameter. A 25 mm window is not a 25 mm usable window.
When pressure and mounting decide the thickness
Pressure windows form their own class, and thickness there is calculated rather than chosen. For a circular plate, bending stress “broadly scales with pressure multiplied by aperture radius squared and divided by thickness squared,” and deflection depends on thickness even more strongly [5]. Double the unsupported diameter at constant thickness and the stress roughly quadruples.
Vacuum viewports make the class visible: they ship as assemblies with seals and mounting parts, because the window is a structural part of the chamber wall [2]. Mounting adds its own loads. A tight retaining ring or a shrinking adhesive can stress a window that handles the pressure differential just fine [5]. The thickness spec, in other words, belongs to the whole assembly - window, seal, and housing together.
How the four dimensions interact
A worked example: a 1064 nm laser chamber window
Suppose you must specify a window for a 1064 nm laser chamber - a hypothetical case, but a typical one. Wavelength admits many substrates: N-BK7, fused silica, and sapphire all transmit at 1064 nm, so this dimension only sets the shortlist.
Material then filters by environment. Flying debris and absorbed laser power tip the choice toward fused silica, whose low expansion holds the surface figure as the part heats. Sapphire’s extra cost buys abrasion resistance the chamber may not need.
Coating comes next and it is narrowband: a V-coat at 1064 nm drives per-surface loss from 4 percent toward zero, which protects both throughput and anything downstream of the back-reflection. Size closes the loop. The beam diameter plus margin sets the clear aperture, and the chamber’s pressure differential - not catalog habit - sets the thickness.
Traps that cross dimensions
Three traps catch engineers who treat the dimensions independently. First, the coating class inherits from the substrate. An uncoated high-index IR window can lose more light than a cheaper coated alternative, so material and coating must be priced as a pair.
Second, thickness interacts with coherence. Extra thickness in a laser beam builds etalon fringes between the two surfaces. The common fix - a wedged window - moves you onto the shape axis, which changes mounting and beam offset.
Third, expansion must match across the assembly. A low-expansion window sealed into a high-expansion housing creates stress under temperature cycling, reproducing the very distortion the substrate choice was meant to avoid. Each dimension narrows the next; none of them decides alone.
FAQ
What is the difference between a UV, visible, and IR optical window?
The band class, and the substrate decides it. A UV window transmits where UV-grade fused silica works, below 200 nm [3]. A visible window is usually N-BK7 or fused silica. An IR window is a crystal or semiconductor such as zinc selenide, germanium, or silicon, chosen for the specific thermal band [2].
Do all optical windows need an AR coating?
No. An uncoated glass window loses about 8 percent of the beam, and plenty of applications tolerate that [7]. High-index infrared substrates are the exception: their surface losses are so large that coating them is effectively mandatory [5].
Which optical window material is the most durable?
Sapphire, by a wide margin. Its Knoop hardness of 2200 trails only diamond among window materials, and it withstands pressure differentials and abrasion that softer substrates do not [3]. The trade-offs are cost and anisotropy - the crystal cut must be specified [5].
How thick should an optical window be?
However thick the load requires. Bending stress scales with pressure times unsupported radius squared over thickness squared, and deflection scales worse [5]. Diameter, pressure differential, and mounting set the number; a catalog default does not.
Can one window cover the whole UV to LWIR range?
No common substrate spans it all. Calcium fluoride reaches from 130 nm to 9.5 µm, covering UV through mid-IR [3]. The long-wave band needs zinc selenide (0.6-21 µm) or germanium (2-16 µm), both blind in the UV [3]. Wide-range FTIR work turns to soft salts like potassium bromide, reaching toward 26 µm at the price of fragility and water sensitivity [3].
Classify in the right order
Optical windows classify by four specification dimensions, and the order is the method: wavelength range first, substrate material second, coating third, size and thickness last. Spectral range kills most substrates immediately. The survivors sort by hardness, expansion, and index. Index then sets what coating is worth buying, and the aperture with the pressure load finishes the job by fixing the geometry.
Used that way, classification is not taxonomy for its own sake - it is a narrowing funnel. Each dimension hands the next a smaller group of genuinely comparable parts. The final choice is between a handful of windows that all work, differing only in price and margins. Walk the dimensions in order on your next part number, and the catalog stops being a catalog.
Shape - flat, wedged, domed - is still a real axis, but it is the last question, not the first. It enters only after the four specification dimensions have decided what the part must be made of, coated with, and sized to.
Sources
- Understanding Optical Windows - Edmund Optics
- Optical Windows - RP Photonics Encyclopedia
- Optical Windows: the Ultimate Guide - Firebird Optics
- ZnSe Optical Components - Sinoptix
- Optical Windows Guide: Materials, Coatings, Specifications and Selection - KKpy
- Fused Silica - RP Photonics Encyclopedia
- Anti-Reflection Coatings - Edmund Optics
- Understanding Optical Specifications - Edmund Optics