The 131 spectra of the standard reference atlas of starlight, hottest at the top, coolest at the bottom. Colour is the wavelength, brightness is the flux measured there, and the dark bands are what that kind of star swallows. Press play. Hydrogen is deepest at A2 and a quarter of that by G2; the calcium K line, absent from the hottest rows, is the deepest band measured anywhere in the atlas by M9; titanium oxide switches on in the last twenty rows and nowhere else. Grey is light outside the window an eye works in — an M9 giant puts 0.2% of its light in there, against 56% for an F0 bright giant. Each row is an average of many real stars of its type, not a photograph of one named star. Tap or hover any row.
Data & source
Data: Pickles
(1998) stellar spectral flux library, UVKLIB set, distributed by STScI in the
CDBS/synphot grid · free to use with citation to Pickles, A. J. 1998, PASP 110, 863
and to STScI; no key, no registration · — spectra,
each 1150–25000 Å on a uniform 5 Å grid · fetched —.
These are not photographs of named stars. Pickles built each
spectrum by averaging and splicing real observed spectra of many stars of the
same type onto one common wavelength grid, so a row is "what an M2 giant looks
like", not "what Betelgeuse looked like on a Tuesday". The atlas covers 131
types across luminosity classes V, IV, III, II and I, mostly at solar
metallicity, with a smaller metal-weak and metal-rich set — 108 solar, 11 weak,
12 rich. It is a reference library, not a survey of the sky, and nothing here
says how common any of these stars are.
The colour. Every wavelength is mapped through an analytic
approximation to the CIE 1931 2° standard observer (Wyman, Sloan & Shirley
2013) into sRGB, taken at full brightness for its hue, and then dimmed by the
flux actually measured at that wavelength — so a dark band on this page is a
real hole in the light, not an annotation drawn on top. Spectral colours mostly
fall outside what a screen can display: those are clamped into the sRGB gamut,
which means the most saturated hues here are the closest a monitor can get and
not the colour itself. Outside 4000–6900 Å the same flux is drawn in grey,
because the standard observer has essentially nothing there and inventing a
colour for the near-ultraviolet or the near-infrared would be inventing data.
The brightness. Each row is scaled to its own brightest point
inside the 3200–9200 Å window drawn here, which throws away absolute flux on
purpose: an M dwarf and a supergiant differ by a factor of millions, and on a
shared scale 120 of these 131 rows would be black. So brightness compares
wavelengths WITHIN a row and nothing between rows. The one number here that
does compare stars is the visible fraction in the tooltip — the share of the
flux the atlas tabulates across its whole 1150–25000 Å range that falls between
4000 and 6900 Å.
The three traces. Each is a band depth measured from these same
samples: one minus the mean flux inside the line, over the mean flux in the
windows flanking it, in parts per thousand. The exact windows are in the table
below. Flanks are a pseudo-continuum, not a true one — in a cool star
every window is full of metal lines — so these are relative depths on a fixed
recipe, comparable down the sequence, and they are not equivalent widths. Two
consequences are visible on the page and are not errors: the Ca II K depth runs
slightly negative in the hottest rows, because there the flanking windows sit on
hydrogen lines and the calcium line does not exist to be deeper than them; and
four of the 524 measurements are not drawn at all, where the star has under 1%
of its peak flux in the windows the recipe needs. Each trace is scaled to its
own maximum across all 131 rows, so their heights are shapes, not a comparison
of strengths; negative values are drawn at the baseline and the true signed
number is in the tooltip.