ASTRONOMY: H. N. RUSSELL 
401 
are typical of small but definite groups, whose variation, though quite 
distinctive, is entirely unpredictable. The spectra of the stars of the 
first of these groups are similar to one another, and unlike anything 
else.^^ Those of the second group are also peculiar, and appear to be 
variable.^^ Both present problems as alluring as they are difi&cult. 
The spectra of other peculiar variables also deserve investigation. 
(e) New stars are usually pretty fully observed while they remain 
bright, but work remains to be done in following at shorter intervals 
the changes during their later stages. The recent work of Adams and 
Pease^^ indicates that they settle down into Wolf-Rayet stars; but, 
according to Miss Cannon,^^ the spectrum of the Nova in Corona, fifty 
years after its outburst, is now of class K. No one seems yet to 
have followed up Hertzsprung's interesting suggestions^ that stars of 
very small absolute luminosity should be investigated for variability. 
Abundant material for a photographic study must exist in the Harvard 
collection. 
4. Knowledge of the distances of the stars is indispensable in the 
solution of many problems. The nearer ones, to a distance of thirty 
parsecs or so, are now accessible to direct measures of parallax, and 
great activity prevails in photographic observation for this purpose, in 
accordance with a wide and well-considered plan of cooperation. 
In my opinion, however, the greatest need in parallax work at 
present is the investigation and elimination of the systematic errors 
which are still present in the best work, as is shown by the too 
frequent appearance of large discordances — sometimes amounting to 
more than O'iOS — between the results of different observers, although 
the probable errors derived from the internal agreement of each ob- 
server's plates are of the order of =^0'i01. The intercomparison of the 
results of various observers for the same stars is hardly a sufficient 
test for the absence of systematic error, especially as all are using nearly 
the same method of observation. The only secure control is afforded 
by observing stars whose parallaxes can be predicted, from other con- 
siderations, with greater accuracy than they can be observed. This 
demands prediction with a probable error not exceeding =tOf005. 
Fortunately, several groups of stars exist for which such prediction is 
possible. The most prominent of these consists of those stars of spec- 
trum B which are between 60° and 120° from the solar apex. If the 
parallaxes of these stars are computed on the assumption that their 
individual proper motions are entirely due to the solar motion, the 
resulting errors will correspond to a probable error of less than one- 
third of the parallaxes themselves — that is, to about =tO'!002. The 
