Positive Electrons hi the Sodium Atom, 277 
No very definite results were obtained, however, and more 
recent experiments show pretty conclusively that the rotatory 
power of most of the absorption lines is confined to wave- 
lengths on one side of the line only. This same action is 
observed in an exaggerated degree by the ultra-violet 
absorption-line of mercury (A, = 2536) which, as Thave shown 
in a previous paper (Astrophys. Journ. July 1907), broadens 
very unsymmetrically . The form of the absorption-curve, and 
the magnetic rotation as shown with the Fresnel rotating 
quartz prisms, is shown in PL XIV. fig. 1, a and b. The 
spectrum obtained by passing white light through the vapour 
placed between crossed polarizing prisms is shown in fig. 1 c, 
the fainter line being rotated 270°. 
The behaviour of mercury vapour will be fully treated in 
a subsequent paper, and for the present we need only remark 
that an absorption-band is possible which only gives an 
appreciable magnetic-rotation for wave-lengths bordering it 
one side. 
This shows us that the lines of the magnetic-rotation 
spectrum would not necessarily appear double, even with the 
highest resolving powers (neglecting rotations larger than 
90°) . Though the lines appear as narrow as arc-lines even 
with a large grating, the magnetized sodium vapour and 
polarizing prism show us that in reality each line embraces 
a narrow range of the spectrum, the individual components 
of which are rotated through very different angles by the 
vapour. 
The experiment which finally showed clearly the nature of 
the rotation was made with a pair of Fresnel quartz-prisms. 
They were much thinner than those usually employed, as it 
was felt that it would be better to work with a single broad 
band of extinction, than a large number of parallel bands. 
The magnetically rotated lines are faint in comparison with 
the continuous spectrum from which they are derived, and it 
is consequently important to have the background upon which 
they are to show up as dark as possible. With a thick 
Fresnel prism we have the continuous spectrum at its full 
intensity traversed by a number of parallel dark bands, 
which correspond to the points on the slit at which the plane 
of polarization is parallel to the plane of extinction (long- 
diagonal) of the analysing nicol, which is placed immediately 
behind the slit. There is in consequence more or less diffused 
light from the grating, which renders the background (the 
dark bands), upon which the rotated lines are to appear, 
much too luminous. To get rid of this effect, the best 
