260 Dr. J. W. Nicholson on the Number of 



Thus we see that the greatest range of variation in n^ ia 

 •09 in 3'67, so that the error in the mean value can hardly 

 exceed about one per cent, at most. The greatest range in 

 % is 37 in 181, or one in five, so that this error can be ten 

 per cent. Thus the value of n is not trustworthy to the same 

 extent as that of n%. 



Silver is the metal which Drude has examined most 

 exhaustively, but the same remarks apply to the others, and 

 the accuracy of ??% is always equally certain, in cases where 

 the observations given are sufficient to test it, whereas it is 

 possible for n in some cases to be incorrect by about 10 per 

 cent, for certain metals on which fewer observations were 

 made. Thus k 2 — v 2 , which we have calculated from Drude's 

 results, is liable to the same percentage error as n 2 . and this 

 can be 20 per cent, in the most unfavourable case, but will 

 not be, for the majority of metals, of any such magnitude. 



It may be shown that the existence of such an error would 

 reverse the sign of K for cadmium, and would nearly do so 

 for tin and magnesium. We notice this point, although in- 

 clining to the belief that the values of K for these metals 

 are a result of sympathetic vibration. Some grounds for this 

 belief will appear in a subsequent paper. 



It is, however, worthy of notice that Drude himself ex- 

 presses some doubt as to the accuracy of the value of n for 

 magnesium, for the specimen used became impure and was 

 difficult to deal with. In the case of cadmium, only one set 

 of experiments appears to have been made, and no comments 

 are given concerning it. Tin is a metal of a pronounced 

 crystalline structure, and this influence may be suggested in 

 its case. The experiments made with it agree very well 

 among themselves, and they seem to the writer to be as 

 accurate as for most metals. Although no statement is made 

 to that effect, it is probable that some precautions were taken 

 against this, for they were in the case of copper, in which 

 results are given both for doubly refracting and for isotropic 

 copper. The calculations of a later paper, which may be 

 mentioned now, will indicate that the constants for tin, and 

 the value of p to which they lead, are in accord with 

 others. 



We have seen, therefore, that the greatest experimental 

 error likely in vk or ?i 2 ^ is ten per cent, at most. That of 

 y/vtc cannot therefore exceed about 4 per cent., and this will 

 be the error in N or p at most, when calculated from the 

 proper velocity distribution. 



Errors in cr have been reduced to a minimum by taking 



