making Experiments upon Elliptic Polarization. ' 243 



spar, give two images always sensibly equal in intensity. But 

 an error which could not be at all detected in this way might 

 produce a very great effect in such experiments as those upon 

 the metals, and, for the purpose of comparison with theory, 

 might render them entirely useless, if in the first method of 

 observing we relied upon one set of observations, taking (sup- 

 pose) the values of & and j3' for the true values of Q and /3 ; or 

 if, in the second method, we contented ourselves with merely 

 measuring the angles y' and /'. 



The necessity of attending to the foregoing rules and re- 

 marks will appear from an examination of the experiments of 

 M. de Senarmont, published in the Annales de Chimie* In 

 these very elaborate experiments, which were made upon light 

 reflected at various incidences from steel and speculum metal, 

 the author followed a plan similar to that which I have adopted, 

 and which, in a general way, I had previously sketched in the 

 Proceedings f of the Academy. There was this difference, how- 

 ever, that he used a plate of mica instead of Fresnel's rhomb. 

 Now as he worked with common white light, the use of the 

 mica plate must have rendered two kinds of errors unavoidable. 

 In the first place, it would be impossible always to take the ob- 

 servations for the same ray of the spectrum; and next, as a 

 consequence of this, the thickness of the plate would be gene- 

 rally inexact for the particular ray to which the observations 

 happened to correspond. If the thickness of the plate were 

 exact for a certain ray, it would be very sensibly inexact even 

 for the neighbouring parts of the spectrum ; and as the part of 

 the spectrum to which the observations belonged was continually 

 changing, the results obtained for different incidences and azi- 

 muths would not be comparable with each other, even though, 

 in each separate case, the error of the plate were allowed for 

 and eliminated. The values of 0, however, as determined by 

 M. de Senarmont, would be correct, so far as this error is con- 

 cerned ; those of /3 alone would be erroneous. For the values 

 of were determined in two ways : by measuring the angles 



* Tom. Lxxiii. pp. 351-358. t VOL. i. p. 159. 



R2 



