of the double refraction in strained glass. 315 



mination of the true position of the fringe is concerned, the 

 probable error is well represented by I'. The range of observa- 

 tion was about 4°, representing a total change of wave-length of 

 2000. An error of 1' will then correspond on an average to 8 

 tenth metres, so that the wave-length of extinction is determined 

 with a proportional error of less than 1 in 600 (taking the mean 

 value \ = 5400). 



Now there exist discrepancies in the results from the same 

 glass, taken on different days, which far exceed this amount. 

 Even in the curves corresponding to a set of observations taken 

 at one time, there are irregularities greater than these probable 

 errors can account for, although they are usually less than the 

 differences between observations of different sets. The only con- 

 clusion possible is that there must be considerable changes in the 

 distribution of stress in the block from day to day. I had noticed 

 these differences already in the earlier experiments (Camb. Phil. 

 Soc. Proceedings, Vol. xn. Pt. i. p. 63), but I had supposed (a) that 

 they were due to sudden jars or shake, so that, if they did occur 

 in a set of observations taken in rapid succession, they would at 

 once reveal themselves by an obvious discontinuity in the results : 

 (b) that as they had not so revealed themselves, the observations 

 of one set, although not comparable with the observations of 

 another, so far as the absolute term was concerned, gave reliable 

 results as regards the relative variation with the wave-length of 

 the quantity C. 



3. Later experiments, however, showed me convincingly, that 

 although this might be true in several cases, there remained a 

 much subtler and more dangerous source of error than this. 



Even when the glass block has been so adjusted that it shows 

 a uniform tint between crossed Nicols for any given total load (in 

 my case I had usually chosen the load giving the tint of passage) 

 it does not remain so for other loads. And this change is not an 

 irregular one, but is, on the contrary, perfectly regular and 

 reversible. The mode of bearing of the glass block upon the 

 compressing planes has a definite dependence on the total load 

 and, after many attempts, I found it practically impossible to 

 ensure that the distribution of stress should remain uniform for 

 any considerable range of load. 



My attention was first called to this when, having observed 

 the very definite black band for first extinction in O 152, I then 

 doubled the load, to observe the band for second extinction. The 

 band was practically invisible, and the cause of this turned out 

 to be, on direct examination, that the stress was now no longer 

 uniform. When the load was reduced, the stress again became 

 uniform. When the block was adjusted so that the stress was 



21—2 



