[PLASKETT] RADIAL VELOCITY DETERMINATION 83 
Collecting in the following table the final values we obtain the 
relative errors for the three dispersions. 
SUMMARY OF PROBABLE ERRORS. 








Errors of Single Errors of Single 
Region. Plates. 
Disp’n.| No. | Mean |——————_-— 
Spectro- | t.m. of Velo- 
graph. per | Plates.| city. | Errors | Total | fetal | Accid’l. | Syst. 
mm. of Accid’] | Accid’l 
Setting. | Errors.| y. 
III L 10.1 24 Poko 049) 0 CNE OMS Ol U7 sO. 4y 
TOR 20.2 DIRES 520275 1.00 0:75 0.32 0.68 
I 33.5 38 | -6.01 1.09 1.41 0.70 0.47 0.52 







We have in the first column the spectrograph employed; in the 
second the dispersion in tenth-metres per millimetre at Hy; in the third 
the number of plates measured, and in the fourth the mean velocity of 
these plates. 
The fifth column contains the average probable error of the estima- 
tion of the coincidences in a single region, while the sixth contains the 
average total accidental error as obtained from the final kilometre 
values of the displacement for each region. 
Under the heading of “ Errors of Single Plates’? we have in the 
seventh column the total error obtained from the mean velocitiesof all the 
plates, and in the eighth the accidental error obtained by dividing the 
total accidental error of a single region by the square root of the number 
of regions, while the ninth column is obtained from the two preceding 
columns by taking the square root of the difference of their squares. 
It is difficult to account for the results obtained, especially in the 
errors of single plates for, as before stated, one would expect the kilo- 
metre values of the probable errors to be inversely proportional to the 
linear dispersion. This is approximately true so far as the errors of 
single regions are concerned and the discrepancy can be satisfactorily 
explained by the greater ease and accuracy in the determination of 
coincidences in the single prism spectrograph owing to the decidedly 
smaller curvature of the spectral lines. But when we come to the total 
errors of a single plate as determined from the measured velocities of 
the plates, we find the errors instead of being in the ratio of 1, 2, and 3 
as we should expect are as 1, 1-1/2, and 1-1/2 approximately. 
So far as III L and III R are concerned it must be remembered 
that, although the linear dispersions are as 1 to 2 the angular disper- 
sions and the resolving powers are equal and the decrease of the ratio 
