450 Lord Rayleigh. On the [Feb. 18, 



individual experiments, would give a better result for the atomic 

 weights than any I could hope to obtain directly. 



In all work of this sort, the errors to be contended with may be 

 classed as either systematic or casual. The latter are eliminated by 

 repetition, and are usually of no importance in the final mean. It is 

 systematic errors that are most to be dreaded. But although directly 

 of but little account, casual errors greatly embarrass a research by 

 rendering difficult and tedious the detection of systematic errors. 

 Thus, in the present case, almost the only source of error that can 

 prejudice the final result is impurity in the gases, especially in the 

 hydrogen. The better the hydrogen, the lighter it will prove ; but 

 the discrimination is blunted by the inevitable errors of weighing. 

 After perhaps a week's work it may become clear that the hydrogen 

 is a little at fault, as happened in one case from penetration of 

 nitrogen between the sealed-in platinum electrodes and the glass of 

 the generator. 



Another difficulty, which affects the presentation of results, turns 

 upon the one-sided character of the errors most to be feared. As 

 has been said, impure hydrogen can only be too heavy, and another 

 important source of error, depending upon imperfect establishment of 

 equilibrium of pressure between the contents of the globe and the 

 external atmosphere, also works one-sidedly in the same direction. 

 The latter source of error is most to be feared immediately after a 

 re-greasing of the tap of the globe. The superfluous grease finds its 

 way into the perforation of the plug, and partially blocks the passage, 

 so that the six minutes usually allowed for the escape of the initial 

 excess of pressure in the globe may become inadequate. Partly 

 from this cause and partly from incomplete washing out of nitrogen 

 from the generator, the first filling of a set was so often found ab- 

 normally heavy that it became a rule in all cases to reject it, From 

 these and other causes, such as accidental leakages not discovered at 

 the time, it was difficult to secure a set of determinations in which 

 the mean really represented the most probable value. At the same 

 time, any arbitrary rejection of individual results must be avoided as 

 far as possible. 



In the present work two objects have been especially kept in view. 

 The first is simplicity upon the chemical side, and the second the use 

 of materials in such a form that the elimination of impurities goes 

 forward in the normal working of the process. When, as in the 

 former determinations, the hydrogen is made from zinc, any im- 

 purity which that material may contain and communicate to the gas 

 cannot be eliminated from the generator; for each experiment 



spheric conditions would remain as 2 : 1 upon indefinite expansion. According to 

 the formula of Van der Waals, a greater change than this in the ratio of volumes 

 is to be expected. 



