APPENDIX. 263 



under consideration, one has its atomic weight unaffected, 

 one is rendered less in accord with the hypothesis, and five 

 approximate more closely than before to even multipfes or 

 multiples half of hydrogen. 



In the later chapters of this work the effect of Dumas' 

 correction is generally less striking. One general state- 

 ment, however, may be made concerning it. Whenever the 

 atomic weight of a metal is calculated from the ratio between 

 its haloid salts and metallic silver, the total effect of Dumas' 

 correction, including the above corrections for the halogens 

 themselves, will be to lovjcr the final result. This point will 

 be further considered presently. Only chlorine, bromine, 

 and iodine have their atomic weights raised by the correc- 

 tion. 



In view of Dumas' correction the question naturally arises 

 as to how far other metals, used in atomic weight researches, 

 may occlude gaseous impurities. For example, when the 

 atomic weight of oxygen is fixed by the synthesis of water 

 over copper oxide, may not the copper occlude appreciable 

 quantities of the hydrogen in which it cools ? If it does, 

 then the apparent weight of metallic copper would be too 

 high, and the atomic weight of oxygen would come out too 

 low. Such an error might possibly account for the differ- 

 ence between 16 and 15.9633 in the atomic weight of oxygen, 

 and it would also increase the atomic weight of coj)per as 

 determined by the same process. At all events, every metal 

 of which the atomic weight has been determined by the re- 

 duction of its compounds in hydrogen, ought to be scrupu- 

 lously investigated with reference to the possible occlusion 

 of gaseous impurities. With all of these metals the effect 

 of such impurities would be to render the apparent atomic 

 weights decidedly too high. 



Although every series of atomic weight determinations 

 must be considered by itself, and weighed on its own merits, 

 it may not be out of place for me just here to point out two 

 general sources of error in addition to the one we have been 

 considering. First, every value after oxygen, with one or 

 two partial exceptions, involves whatever error may attach 



