of the double refraction in strained glass. 59 



scale alongside the continuous spectra of the carbons, in which 

 the dark bands were observed. 



The first step was to identify a large number of well-defined, 

 easily recognisable, lines in the arc spectrum. This was done by 

 comparison with the solar spectrum, from which a number of lines 

 were first of all identified. A curve was then obtained, connect- 

 ing wave-lengths with degrees read off on the spectroscope circle. 

 By means of this curve the approximate wave-lengths of other 

 lines were obtained and characteristic groups identified from 

 Watts' and Rowland's lists of spectra. In this way a very great 

 number of lines were found and a very accurate curve con- 

 necting the spectroscope readings with the wave-lengths deduced. 



The spectroscope and all the arrangements described in Fig. 1 

 were, throughout all this latter part of the work and the experi- 

 ments which followed, kept fixed. The block B had been put into 

 place and the Nicols had been accurately crossed. The apparatus 

 was then touched only for loading and unloading the straining 

 apparatus ; when it was desired to take an observation of the 

 spectrum, the block B was slightly loaded, so as to let the light 

 pass through. Thus the mapping of the spectrum was carried out 

 with all the apparatus under precisely the same conditions as 

 when the bands were observed. 



The value of these precautions will become apparent when it 

 is explained that the bright line spectrum was found inconvenient 

 for direct reference, as it was not by any means so accurate to 

 estimate the position of a diffuse black band with regard to two 

 lines (which would often be inconveniently placed) in a bright 

 line spectrum below it, as to place a cross-wire upon the middle 

 of the band. 



Accordingly, once the comparison spectrum had been mapped, 

 the method adopted was simply to take the circle readings of the 

 black bands and to deduce the corresponding wave-length from 

 the curve of wave-lengths to deviations for the spectroscope. In 

 order, however, to eliminate any small shifts due to temperature 

 &c, which might occur from day to day, the circle readings of two 

 well-defined lines (D and E of the solar spectrum) were taken on 

 each occasion and the difference of the actual readings of these 

 two lines from the readings corresponding to them in the initial 

 diagram (which difference was in all cases small) was treated as an 

 index error. 



5. In every case, only the black bands of the first order, that is, 

 those corresponding to a relative retardation of one whole wave- 

 length, were observed. We had therefore the formula 



X = -P, 



a 



