of the double refraction in strained glass. 63 



fitted by a straight line parallel to the dotted line, but differing 

 slightly in position. 



The explanation of this is probably that between the first and the 

 second set of observations, which were not taken on the same day, 

 the straining apparatus suffered some slight jar, which must have 

 altered the distribution of stress in the block B, raising or lower- 

 ing by a small amount the value of the stress in the. particular 

 region through which the light passed. 



For this reason a complete set of observations was in all cases 

 taken as nearly as possible at the same time, and only observations 

 of the same group are, in strictness, comparable. Nevertheless the 

 results of July 9 and 11 agree so well that I have felt justified in 

 grouping them together. 



It may be noticed here that, for the reason mentioned above, 

 although the present is a delicate method for detecting variations 

 in the stress-optical coefficient (the stress, whatever the distri- 

 bution, being always proportional to the total load, so long as the 

 mode of application of the latter is unchanged), it is a very bad 

 one for obtaining its absolute value. 



For, according as the stress in the middle of the block is 

 greater or less than the mean value calculated from the total load, 

 the calculated coefficient C will be greater or less than its true 

 value. 



No very great importance, therefore, should be attached to the 

 absolute values of G given in Tables I. — III. It will be seen, on 

 comparing with the table on p. 491 of the paper quoted (Camb. 

 Phil. Proc, Vol. XI. Pt. VI.), that those values are considerably 

 less (about 10 per cent.) than those found for the two beams used 

 in the Cambridge experiments 1 . 



This of course may be due to the fact that glasses, outwardly 

 similar, may differ in their stress-optical properties, the glass used 

 in these experiments not having been cut from the identical 

 beams employed in the Cambridge observations. 



It appears far more likely, however, that this discrepancy in 

 the absolute value is due to the fact that the stress in the centre 

 of the block was less than its mean value, a result obtained 

 theoretically by the author when considering an elastic cylinder 

 compressed between lead sheets, which force the ends to expand 

 (" On the equilibrium of circular cylinders under certain practical 

 systems of load," Phil. Trans. A, Vol. 198, pp. 147—233). 



If we bear in mind these causes of error, to which the determi- 

 nation of the absolute value is subject, we see that, as far as the 

 relative values are concerned, the observations of Fig. 3 indicate 

 quite a good agreement between the three series. 



1 I regret that, owing to an error in copying, a 10 -6 appears multiplying the 

 figures in the last column of the table referred to. It should be 10 -7 . 



