320 SIR, F. W. DYSON, PROF. A. S. EDDINGTON AND MR. C. DAVIDSON ON A 
The sign of the results shows that the scale of the photographs is larger at Principe 
than at Oxford ; in fact the focus must have been set about 1 • 2 mm. further out 
(apart from any change of length compensated by expansion of the photographic plates). 
As the error in focussing was probably not more than 0*5 mm., the greater part of this 
shift must be due to the focal length of the lens combination increasing ^vith temperature 
more rapidly tlian the linear expansion of the glass. 
If the only difference were a change of focal length, we should have a' = e . There 
is a fairly strong indication that e' is greater than a' . This is no doubt due to a change 
in the definition caused by the coelostat mirror or by a shift of the object-glass lenses 
on the journey ; and, as it will presumably affect the eclipse plates in the same way, 
it is best to adopt the values of a' and e' as determined, rather than to take a mean. 
In so doing we shall at any rate not exaggerate the displacement, which depends mainly 
on the y-measures and is reduced by adopting too large a value of e'.* 
The difference h' — d' merely gives the relative orientation of the two plates as 
placed face to face. The sum h' -f- d' practically vanishes, as it should do. However, 
for consistency we adopt the small value found. 
From the internal discordances of our determination of e' (the most important of 
these constants) the probable error of the mean is ±2*1. This, as shown later, vdll 
cause a probable error of our final determination of the deflection, reduced to the limb 
of the sun, of amount ±0"-14, affecting all determinations systematically. Errors 
in the other constants have much smaller influence. 
The Eclifse Plates. 
34. The eclipse plates from K to S show no star images. After that the cloud 
lightened somewhat, and some images appear on the remaining plates. The sky was 
never clear and nothing fainter than 5'-5 is shown. The cloud was variable in different 
parts of the plate, so that the brightness of the images varies erratically and the diffusion 
is also variable. 
In order to obtain results of any weight the stars 4 and 3 (kj and /C 2 Tauri), which 
theoretically should be strongly displaced, must be shown. They appear on all plates 
from T to Z, and being near the centre of the field have good images. They are relatively 
rather faint on plate U, but are bright on the other plates. The appearance of the 
remaining stars is as follows :— 
Plate T. 6 bright ; 10 faint. 
Plate U. 6, 10 very bright ; 11 faint. 
Plate V. 6 bright ; 10 fair. 
Plate W. 5, 6 good ; 10 diffused. 
Plate X. 5, 6, 11 good. 
Plate Y. 5, 6, 11 faint, diffused ; 12 very faint. 
Plate Z. 5, 6, 11 faint, diffused. 
* It happens that it is also reduced, but to a less extent, by using too small a value of a’. 
