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



corresponding to this particular wave-length. By sufficiently 

 increasing the brightness of the light, this dark band might be 

 made sufficiently sharply defined to allow of accurate measurement. 



This was, as a matter of fact, found to be the case : no very 

 definite advantage, however, was gained by increasing very greatly 

 the brightness of the light, because it was not found possible to 

 obtain perfect uniformity and perfect verticality of the stress in 

 the block B, so that some small fraction of the light managed to 

 get through, even when theoretically there should have been 

 absolute extinction. Hence the band was not absolutely black and 

 no increase in definition resulted from increasing the brightness of 

 the light beyond a certain limit. 



Nevertheless, well-defined dark bands were obtained and these 

 allowed of being measured with considerable exactness. The 

 spectroscope was provided with a vernier V and a reading lens R 

 by means of which angles could be read up to |-' of arc. The 

 probable error of the position of a band varied considerably accord- 

 ing to the part of the spectrum observed. In the yellow and 

 green parts of the spectrum, where the background was brighter 

 and the lines more sharply denned, the probable error, as deduced 

 from differences of readings, was under 1'. In the blue and violet, 

 however, where the bands were diffuse and the background faint, 

 the readings were far more uncertain. 



The probable error also varied considerably from day to day, 

 as was natural. On four different days the following probable 

 errors were calculated from differences of readings in all parts of 

 the spectrum: l'*49, 2H8, 0'-60, 3'-15. The mean of these, l'-93, 

 may be taken as a fair representation of the average error. The 

 range of angle being about 8°, this corresponded to about 4 per 

 1000 of the range. 



The w r eight could be found easily to the nearest gramme 

 and this gave an accuracy of 1/10,000, or even greater accuracy, 

 the weight used exceeding 10 kilogrammes in each case. Hence 

 in calculating the value of C, the error of 4/1000 of the whole 

 range (X red — \ violet) would amount to about 10 tenth-metres. 

 This for a wave-length of 5000 would give an error of 1/500 in 

 the wave-length of the light quenched, and since 



, GP 



nX = , 



a 



the error in C would also be 1/500. This is nearly 10 times the 

 accuracy of the observations of the earlier experiments (Camb. 

 Phil. Proc, loc. tit.). 



4. In the present experiments no special comparison spec- 

 trum was used, the spectrum of the arc itself forming a most 

 beautiful natural comparison spectrum, spread out like a luminous 



