228 NATURE 
[APRIL 30, 1914 
cipal differences in stellar spectra, however they may 
originate, arise in the main from variations in a 
single physical condition in the stellar atmospheres. 
This follows at once from the linearity of the series. 
If the spectra depended, to a comparable degree, on 
two independently variable conditions, we should 
expect that we would be obliged to represent their 
relations, not by points on a line, but by points scat- 
tered over an area. The minor differences which are 
usually described as “‘ peculiarities’? may well repre- 
sent the effects of other physical conditions than the 
controlling one. 
The first great problem of stellar spectroscopy is 
the identification of this predominant cause of the 
spectral differences. The hypothesis which suggested 
itself immediately upon the first studies of stellar 
spectra was that the differences arose from variations 
in the chemica! composition of the stars. Our know- 
ledge of this composition is now very extensive. 
Almost every line tn the spectra of all the principal 
classes can be produced in the laboratory, and the 
evidence so secured regarding the uniformity of nature 
is probably the most impressive in existence. The 
lines of certain elements are indeed characteristic 
of particular spectral classes; those of helium, for 
instance, appear only in Class B, and form its most 
distinctive characteristic. But negative conclusions 
are proverbially unsafe. The integrated spectrum of 
the sun shows no evidence whatever of helium, but 
in that of the chromosphere it is exceedingly con- 
spicuous. Were it not for the fact that we are near 
this one star of Class G, and can study it in detail, 
we might have erroneously concluded that helium was 
confined to the “helium stars.” There are other 
cogent arguments against this hypothesis. For 
example, the members of a star-cluster, which are all 
moving together, and presumably have a common 
origin, and even the physically connected components 
of many double stars, may have spectra of very 
different types, and it is very hard to see how, in 
such a case, all the helium and most of the hydrogen 
could have collected in one star, and practically all 
the metals in the other. A further argument—and 
to the speaker a very convincing one—is that it is 
almost unbelievable that differences of chemical com- 
position should reduce to a function of a single vari- 
able, and give rise to the observed linear series of 
spectral types. 
I need not detain you with the recital of the steps 
by which astrophysicists have become generally con- 
vinced that the main cause of the differences of the 
spectral classes is difference of temperature of the 
stellar atmospheres. There is time only to review 
some of the most important evidence which, converg- 
ing from several quarters, affords apparently a secure 
basis for this belief. ; 
The first argument is based upon the behaviour of 
the spectral lines themselves. To appreciate its full 
force, one must familiarise himself with a multitude 
of details. A typica' instance is that of the heavy 
bands in the region of longer wave-length, which are 
the most characteristic fea‘ure of spectra of Class M, 
appear faintly in Class Ks, and are absent in Class K 
and all those higher in the series. Fowler has shown ! 
that these bands are perfectly reproduced in the spec- 
trum of the outer flame of an electric arc charged 
With some compound of titanium, while the spectrum 
of the core of the arc, though showing conspicuously 
the bright lines of titanium, does not contain the 
bands. Here we are evidently dealing with some 
compound—perhaps titanium oxide—the vapour of 
which is present in the relatively cool flame of the 
arc, and emits a spectrum of the banded type, char- 
1 Proc. Roy. Soc., vol. Ixxii., pp. 219-225, 1904. 
NO.-2322,"yOr-nos| 
acteristic of compounds, while in the hotter core it is 
dissociated, and only the lines of the metal are seen. 
There seems then to be no escape from the conclusion 
that the atmospheres of stars of Class M are cool 
enough to permit the existence of this compound, and 
hence cooler than the core of the arc, and that the 
temperature of its dissociation is approached in 
Class K5, and surpassed in Class K. In general, 
those metallic lines which are relatively strong in the 
spectra produced in the oxyhydrogen flame or the 
electric furnace are also strong in spectra of Classes 
M and K; the lines most prominent in Class G are the 
typical arc lines; and the relatively few metallic lines 
which persist into Classes A and B are those which 
appear exclusively, or with greatly enhanced intensity, 
in the spark spectra of the laboratory. 
The second line of evidence is afforded by the dis- 
tribution of intensity in the continuous background 
of the spectra, the differences of which from type to 
type are obvious to the eye as differences in. the colour 
of the stars. This characteristic is fortunately capable 
of accurate measurement. For the brighter stars, 
spectro-photometric comparisons may be made with a 
terrestrial light-source the energy curve of which is 
known, as has been done visually by Wilsing and 
Scheiner,* and photographically by Rosenberg.? Much 
fainter stars may be reached by the comparison of 
their brightness as measured visually (or on iso- 
chromatic plates with a suitable colour-screen), and 
photographically on ordinary plates. The ‘colour- 
index ’’ so obtained, which expresses, in stellar mag- 
nitudes, the relative photographic brightness of stars 
of equal visual brightness, is found to be very inti- 
mately related to the spectral type, the differences 
within each spectral class being scarcely greater than 
the errors of observation. The results of King,* Park- 
hurst,? and Schwarzschild,® working with different 
instruments and on stars of very different brightness, 
are in excellent agreement, as is shown in Table I. 
The near approach to equality among the differences 
in colour-index from class to class is very remarkable, 
when it is considered that these types were picked 
out somewhat arbitrarily according to the general 
appearance of the photographic spectra. The judg- 
ment of the Harvard observers in selecting the really 
important points of difference was evidently very good. 
TABLE I. 
: Colour-index Temperature 
«Spectrum King Parkhurst Schwarzschild : 
Bo — 0-32 20,000 
Bs —0-17 —0-21 — 0:20 14,000 
Ao 0:00 0:00 0-00 11,000 
A5 0-19 0:23 0:20 9,000 
Fo 0:30 0°43 0-40 7,500 
F5 0-42 0:65 0-60 6,000 
Go 0-72 0-86 0:84 5,000 
Gs 0:98 1-07 I-IO 4,500 
Ko I-10 1-30 1-35 4,200 
K5 1-62 I-51 1-80 3,200 
M 1-62 1-68 3,100 
N 2: 2,300 
If the spectral sensitiveness of the plates used in 
such investigations has been determined (as Park- 
hurst has done) it is possible to calculate the tempera-. 
ture at which a black-body would emit light of the 
same colour as that observed; and similar calculations 
can be made, with greater accuracy, from the spectro- 
photometric data. The last column of Table I. gives 
the effective temperatures thus derived (based mainly 
on the work of Wilsing and Scheiner). The absolute 
2 Potsdam Publications, vol. xix., part 1. 3% A.N., 4628, 1913. 
4 Harvard Annals, vol. lix., p. 179. 
5 Astrophys. Jour., vol. xxxvi., p. 218, 1912. 
6 Gottingen Aktinometrie, Teil B, p. 19. 
