Wave-length of Light, 



271 



To eliminate any errors which might be due to imperfec- 

 tions of figure in the gratings, they were used in all the four 

 possible positions. No such error, however, became apparent 

 either from critical examination of the gratings themselves, or 

 from the results obtained in the different positions. 



Observations with grating I. were begun late in October, 

 1885, and occupied the clear days for a month. Forty-eight 

 series of measurements were made, and the agreement be- 

 tween them was very satisfactory. After correcting for tem- 

 perature, thirty-six of the number fell within a range of three 

 seconds, and the rest were clustered closely about them. 

 Observations on the various days were as follows: — 



Date. 



Number of 

 Series. 





Angle. 



Oct, 19 . . . 



1 



45° 



V 47"-2 



20 . . . 



1 



45 



1 48 -4 



22 . . . 



2 



45 



1 48 -2 



23 . . . 



1 



45 



1 49 -8 



26 . . 



4 



45 



1 49 '3 



27 . . 



3 



45 



1 48 -2 



31 . . 



1 



45 



1 50 -1 



Nov. 3 . . 



1 



45 



1 48 -6 



4 . . 



3 



45 



1 47 -4 



5 . . 



2 



45 



1 47 -9 



10 . . 



4 



45 



1 47 -8 



11 . . 



6 



45 



1 49 -7 



16 . . 



8 



45 



1 48 -2 



17 . . 



5 



45 



1 47 -5 



20 . . 



6 



45 



1 47 -5 



All the above were in the third spectrum, to w T hich mea- 

 surements were in the main confined, as in it the definition 

 was particularly good ; and it being the highest order which 

 could be conveniently observed, an error in the angle would 

 produce the minimum effect. The spectra on both sides of the 

 slit were about equal in brilliancy and definition. 



The observations were weighted as nearly as possible accord- 

 ing to the favourable or unfavourable conditions under which 

 they were made ; and when finally combined, gave as the value 

 of the angle of deviation for grating I. : — 



</> = 45°l / 48 // -24 + // 'll. 



The above probable error is equivalent to a little less than 

 one part in a million, and can introduce no sensible error into 

 the resulting wave-length. 



Other work intervened, and the measurements with grating 



