( 356 ) 



id say llic leasl hears a very striking resemblance to the one deduced 

 from theory by Voigt. 



In both cases, the theoretical one and thai following from our 

 experiments, there is a difference of the distances between the central 

 line and the two miter components, in this sense that the component 

 towards the red is nearer to the middle line than that towards the 

 violet, just as predicted by theory. 



There exists also an asymmetry of the intensities of the outer 

 components in the sense indicated by theory. 



An inspection e. g. of the original negative of which Fig. 1 is a 

 nine-fold reproduction, or of the reproduction Fig. 1, or better of 

 reprints on photographic paper of the 29-fold enlargement given in 

 Fig. 2 or even of that figure reveals the existence of a very small 

 asymmetry of intensity. This is perhaps most clearly seen by looking 

 al the figure from a not loo small distance, covering the central line 

 wiih a small strip of paper. No trace of asymmetry can be seen in 

 the triplet of line 5770, see also the enlargements Fig. 4 and Fig. 5 

 of the middle and outer parts of the right of Fig. I. 



On the other hand there seems to be a difference between theory 

 and observation in this respect, that the amount of asymmetry appears 

 to be not constant. The table of § 5 and the graphical representation 

 clearly indicate that when the magnetic force decreases from 30000 

 to 15000 Gauss the asymmetry also is nearly halved. '). 



An error of an amount sufficient to bring a single point of the 

 upper line on the dotted one is not absolutely excluded (see § 9). 

 [For the right part of the diagram the error ought to be three times 

 the probable error of one single of' the principal values (see § 5) and 

 would happen therefore on the average in one out of every twenty 

 three cases]. 



We have however reasonable security against a combination of 

 errors which would move all the points of the full line to the 

 dotted one. 



Of course we cannot deduce from the now determined part of 

 the upper line whether or not it will approach asymptotically to a 

 finite distance of the lower one. 



9. We may now consider the question as to (he best fitting 

 straight lines to our two systems of points. 



.Measuring the divergencies at right angles to the line the best tit 

 will be obtained if we make the sum of the squares of the perpen- 



l ) An excellent series of measurements made after the writing of this article 

 gives a somewhat lower rate of decrease, the mean value of the asymmetry 

 being the same. 



