2. Reading the 1931 chart and its limits

The curved edge is the spectral locus

The spectrum locus, commonly called the spectral locus, traces monochromatic chromaticities. A finite-bandwidth LED or OLED generally lies inside this boundary because its spectrum combines neighboring wavelengths. Narrower bandwidth tends to approach the locus, but a point’s distance from the boundary does not uniquely determine its bandwidth.

CIE’s retained 1 nm color-matching dataset covers 360–830 nm. This is a tabulation range, not a sharp physiological visibility cutoff. Wavelength labels are dense near some parts of the curve and sparse near others. The eye’s sensitivity, available radiance and viewing conditions matter. A spectrum locus drawn from rounded or sparse data is a numerical approximation.

The line of purples closes the boundary

CIE describes the purple boundary using mixtures of monochromatic stimuli at approximately 380 and 780 nm. The full-range tabulated diagram closes its locus at the retained extreme wavelengths; endpoint details depend on the adopted data range. These practical conventions should not be mistaken for a new band of spectral wavelengths.

The ideal boundary mixtures are limiting cases. Ordinary displays use finite-bandwidth primaries and reproduce only a subset of purple chromaticities. The word purple also covers interior colors, not only boundary points. A magenta screen pixel is usually a mixture of the display’s red and blue primaries; it need not lie on the outer purple boundary. Equal powers do not in general put the point at the geometric midpoint.

Physical limits are stricter than x + y ≤ 1

The familiar horseshoe and its closing purple boundary show the practical colorimetric limit for nonnegative spectra under the chosen observer. Computationally, the convex hull of the sampled spectral chromaticities is a useful physical-region approximation, also used by the calculator. Tiny local irregularities of tabulated data should not be interpreted as new producible colors.

The axes themselves are coordinate limits, not spectral limits. A triangle with vertices (0,0), (1,0) and (0,1) describes nonnegative normalized XYZ; many points in it still cannot be generated by light. Imaginary XYZ primaries are mathematical basis vectors. A manufacturer’s bounding rectangle is an allowed coordinate envelope, not proof that every corner can occur.

The marked example is (0.01, 0.90) in xy, which transforms to approximately (0.00290, 0.58781) in u′v′. Although x, y and z=0.09 are all positive, no nonnegative light spectrum can generate it for this observer. The colored interior is the physically realizable chromaticity region; its on-screen colors are illustrative approximations.

What neither chart can tell you

Chromaticity is not a full color-appearance model. Adaptation, surround, luminance, contrast, field size and observer differences influence appearance. Two samples can match in xy at one condition and differ perceptually at another. Normalized plots discard intensity; dimming an invariant spectrum leaves its point unchanged, although actual devices may shift spectrum when dimmed.

At X=Y=Z=0, both xy and u′v′ formulas produce 0/0. Near black, measurement noise and stray light can make reported chromaticity unstable. A browser’s plotted fill uses an sRGB approximation with clipping or normalization; it cannot demonstrate the physical appearance of an out-of-gamut green, a radiance level or NVIS performance.