314 Mr Filon, On the variation with the wave-length 



Table I. 



N.B. The materials in brackets are only used in very small 

 quantities (1 — 2 per cent.). 



2. The method of conducting the experiments was practically 

 the same as the one described in the previous paper referred to. 

 One or two alterations of detail, however, were made. Thus, after 

 the first two sets of experiments on 103 it was decided to 

 remove one of the prisms in the spectroscope, as the additional 

 accuracy in reading the circle was more than compensated by 

 the greater diffuseness of the band, which practically rendered 

 observations in the violet useless, the very broad hazy band being 

 almost undistinguishable from the faint background. The re- 

 moval of the one prism narrowed the band, and by decreasing 

 the absorption due to the large thicknesses of glass traversed, 

 considerably brightened the surrounding field in the violet. 



With regard to the errors of measurement made under the 

 new conditions, I took the probable errors on four different occa- 

 sions, when the conditions were good, moderate, and poor. The 

 probable errors were found by taking the means of differences of 

 two readings which should have been the same and dividing by 

 V2. They were found to be /- 6, l'l, l' - l, l'*4 respectively, and 

 these were found from observations in all parts of the spectrum, 

 including the violet, in which the discordances are always greater. 

 The accuracy in the brighter parts of the spectrum was probably 

 better. Nevertheless, we may assume that, as far as the deter- 



* I cannot help thinking there must be some error in the composition stated 

 for this glass, as, comparing with the one next in the list, it seems unlikely that a 

 small percentage of K 2 should cause such a large difference in specific gravity. 

 This however does not affect the present paper, for no results, as will be seen, 

 could be obtained with O 198, 



