174 TRANSACTIONS OF THE [APR. 1, 
strongly that in the case of uranium glass, at least, the phenome- 
non is one of simple refraction at the surface; but that the 
WHOLE of the emitted light undergoes the refraction process. 
dG 
EXPERIMENTS UPON PLATINUM. 
It is evident that no such comparisons as those just made for 
uranium glass could be made for the case of incandescent metals, 
unless, in the first place, the surface experimented upon could 
be assumed to be a perfectly definite, non-diffusing surface. The 
chief source of difficulty in the work upon platiuum was to ful- 
fill this condition. 
It was found, after considerable work had been done upon 
platinum, that continual heating roughened the surface to a 
slight degree, and changed the amount of polarization. The re- 
sults of several sets of laborious observations upon platinum 
were discarded altogether, because they were found to be erron- 
eous from this cause. However, the change is so gradual that a 
well polished platinum surface may be heated to incandescence 
for several minutes without showing any perceptible change in 
character. The rapidity of the change could be delicately ob- 
served by viewing the surface at a large angle of incidence by 
means of the polarimeter. For a period of two or three minutes 
no change was perceptible in the equality of the images, but for 
much longer periods of heating the slow blistering of the surface 
began to be manifest in the disturbance of the equality of the 
images. Hence, in order to avoid this error, the surface of the 
platinum was carefully polished with rouge after every set of 
readings for a given angle. 
A second slight source of error in the observations upon plat- 
inum was the lack of exact horizontality in the surface examined. 
The attempt was made to avoid this error by rotating the in- 
strument through 90° according to the suggestion of Cornu. 
This brought the extraordinary image either above or below the 
ordinary ; hence, when the angle of emergence was very large, the 
two images corresponded to points on the surface at a consider- 
able distance from each other, 
as shown in the figure re 
This introduced the likeli- 
hood of a much greater error than that due to a slight error in 
horizontality. The incandescent platinum was therefore rend- 
ered as nearly horizontal as possible by comparison with care- 
fully levelled reference planes placed in the immediate vicinity. 
The adjustment could thus be easily made to within one degree. 
