The Anomalous Dispersion of Sodium Vapour. 163 



diamond. In this way a scale of wave-lengths was secured. For the 

 determination of refractive indices we require, however, the angular 

 deviation of the different rays. This can be determined if we know 

 the focal length of the lens forming the image of the spectrum on the slit 

 of the spectrometer, and the actual deviations measured in millimetres 

 of the images of the first slit formed by light of various wave-length at 

 this point. To secure a record of this a glass plate, ruled with half- 

 millimetre lines, was placed over the slit of the spectrometer, and the 

 position of the lines on the glass plate in the focus of the large lens 

 recorded with the writing diamond. This scale enabled the recorded 

 deviations to be reduced to the actual deviations as they existed on the 

 slit of the instrument. 



The dispersion curve obtained in this manner, with the half-milli- 

 metre marks, is reproduced on the same scale as the original in fig. 4, 

 the prismatic spectrum having been converted into a normal spectrum. 

 It will be seen that for all wave-lengths shorter than those of the D 

 lines, the refractive index is less than one, while for the rest of the 

 spectrum it is greater than one. This is the only case that I know of 

 in which we have a medium, beautifully transparent even in consider- 

 able thicknesses, in which light travels faster than in a vacuum. 



Fio. 4. 



DLJnes. 



Wave-Length scate. 



Line Of Zero ^^^ 58756SSS4a&6l SO494a47464i444i4Z4l 40393037 



deviation- 



-HaLfmiLLimetre 

 - masks. 



To determine the refractive indices we require the angle of the 

 sodium prism, and here we encounter a grave difficulty, for it appears 

 to be quite impossible to determine this directly. As I halve said 

 before, we are dealing with the optical equivalent of a prism, namely, 

 a non-homogeneous medium, in which the vertical wave-front is 

 retarded or accelerated in a progressively increasing amount as we 

 pass from its upper to its lower edge. The equivalent prism appears 

 from experiment to be bounded by concave instead of plane surfaces, 

 in other words the effective angle is greater near the bottom of the 

 tube than at the top. This can be shown by screening off different 

 levels. 



The lower part of the tube acting alone gives a much greater 

 deviation than the upper. In practice I have found it best to screen 



