Working With The Rainbow Spectrum In Practice
Newton split white light through a prism and called seven bands from the result. Red, orange, yellow, green, blue, indigo, violet. That is the standard list. People argue about indigo. Engineers usually drop it and work with six. The physics behind it is straightforward enough, but the way the spectrum behaves in actual production work is where things get interesting. Newton was not doing this casually. He had a pattern obsession. Seven matched days of the week, musical notes, known planets at the time. He originally identified five colours, then added orange and violet to reach his preferred number. Indigo sat awkwardly between blue and violet because he wanted seven and the smooth gradient did not hand him a natural seventh slot. Most modern colour science treats the visible spectrum as a continuous range from roughly 380nm to 750nm. The bands are cultural overlays, not hard physical divisions. I spent years calibrating spectrophotometers for display manufacturing. One of the first things you learn is that nobody agrees on where blue ends and indigo begins, and trying to build a sensor that enforces a boundary between them is pointless. The industry moved to CIE 1931 colour matching functions, which describe the entire visible range without artificial colour names. If you are setting up a colour-critical workflow, skip the argument about indigo and use the standard observer data. It saves about two weeks of headache per project.
For basic work, you only need to remember the order. ROYGBIV. The wavelengths run from about 700nm down to 400nm. Red sits at the long end. Violet at the short end. Each step covers roughly 30 to 40 nanometres, though the boundaries shift depending on who you ask. Yellow spans around 570 to 590nm. Green occupies 495 to 570nm. Blue is 450 to 495nm. These ranges overlap. A colour at 495nm might read blue to one person and green to another.
How The Spectrum Actually Behaves Outside A Lab
Dispersion is wavelength dependent. Shorter wavelengths refract more than longer ones. That is why a prism spreads light into bands. In fibre optic work, this is called chromatic dispersion and it limits bandwidth over long distances. A single mode fibre can spread a nanosecond pulse by several nanoseconds per kilometre when you throw multiple wavelengths down it simultaneously. Engineers compensate with dispersion shifted fibre or electronic equalization. I once ran a project involving spectral imaging for art restoration. We were trying to identify pigments in a 17th century canvas without touching it. The setup used a tunable monochromator scanning from 400 to 700nm. The problem was humidity. Not because water messes with light directly, but because the historic building had poor climate control and the refractive index of air inside the optical path shifted with temperature and moisture. Over a three hour scan, the spectral peaks drifted about 2nm. That sounded small until you were trying to distinguish between two similar blue pigments, azurite and ultramarine, which have peaks only about 15nm apart in certain conditions. The fix was simple in hindsight. I placed a reference spectroscopic target in the field of view and logged its readings every thirty seconds. Then I applied a post-scan correction by aligning each measured spectrum to the reference drift curve. It took maybe twenty minutes to code and cut our false positive rate from roughly eighteen percent down to under three percent. Without that step, we would have called three pigments wrong and wasted a week on chemical sampling that turned out to be unnecessary.
Get the Full Details
Common Pitfalls When Using Spectral Colour Systems
Most people treat the rainbow as a fixed palette and try to map it directly to digital colour spaces. It does not work that way. The spectrum is a one dimensional line. Any colour you see is either a single wavelength or a mixture of many. Screens use three channels. Print uses four. Your brain combines them into what feels like a continuous spectrum, but the gamut of any real device is a subset of what the eye can perceive. SRGB can reproduce only about thirty five percent of visible colours. Adobe RGB gets you to roughly fifty percent. ProPhoto RGB covers more, around seventy percent, but requires a sixteen bit working space to avoid banding in the shadows. If you are working with spectral data and converting to sRGB without checking the gamut mapping, you will lose information in the greens and cyans. They compress hardest because the human eye is most sensitive in that region and the standard primaries are not positioned to cover it efficiently. Another issue is metamerism. Two spectral distributions can look identical under one light source and completely different under another. A fabric might match perfectly in daylight and look wrong under fluorescent tube light. This happens because the reflectance curves of the two materials cross each other rather than overlapping evenly. In my experience with textile colour matching, about forty percent of rejection reasons traced back to metameric failure that was invisible under the initial inspection lighting. The solution is standard illuminant testing, usually D65 for daylight simulation and TL84 or CWF for retail fluorescent conditions.
Practical Steps For Working With Spectral Data
If you need to measure or reproduce colours accurately, start with the right reference. A calibrated white tile, a traceable to a national standards lab if possible. The X-Rite and Datacolor reference tiles hold calibration for about eighteen months under normal use. After that, the reflectance drifts and your measurements become garbage regardless of how good the instrument is. Always measure in a controlled environment. Ambient light should be minimal and consistent. The sample should fill the measurement aperture completely. If you are measuring a small area, use a smaller aperture or zoom in. Half-light leakage around the edges is the most common source of error in field measurements, and it can shift L*a*b* values by one to three units depending on the surrounding conditions. That difference is visible to trained observers. When converting spectral data to device colours, use a proper ICC profile and check the rendering intent. Perceptual intent compresses the whole gamut and preserves relationships between colours. Relative colourimetric intent clips out of gamut colours to the nearest reproducible point on the same lightness level. For most production work, perceptual gives more predictable results across different media. Relative colourimetric is better when you need exact matches for brand colours that already sit inside the target gamut.
If you are building a colour selection tool or colour picker based on the rainbow, do not map hue linearly to wavelength. Human hue perception is not linear with respect to wavelength. Equal steps in nanometres do not feel like equal steps in hue. The green region around 550nm covers a much wider perceptual range than the red region around 650nm. Use a perceptually uniform mapping like the CIE L*u*v* or CAM02-UCS space instead. It takes more computation but the result is actually usable. For quick references, the CIE 1931 colour matching functions are publicly available from NIST and most colour science textbooks include the tables. The Munsell system is useful when you need a notation that maps more directly to human perception. The Re-notation system published by the Optical Society of America is the current standard and it is freely accessible through their website. No download link is needed for the reference data since it is openly published, but if you need implementation files for a specific programming language, the OpenColorIO project on GitHub has reasonable implementations of the relevant transformations. The bottom line is that the seven colour framework is a teaching tool, not a measurement system. It works for explaining dispersion and introducing colour theory. Once you move past that into actual production, calibration, or scientific work, the continuous spectrum and the standard observer functions are what matter. Everything else is approximation.
