DETAILS OF ANALYSIS. 53 



The conflicting outcome of all these carefully conducted experi- 

 ments thus still remained not easily explicable. Their value for the 

 atomic weight of sodium differed by over o.i per cent from Stas's 

 result, and furthermore, as has just been said, the comparison of sodic 

 chloride with silver yielded a higher result, 23.027, than that found 

 by comparison with argentic chloride, 23.017. Both these discrepan- 

 cies required explanation before the results could be accepted without 

 reserve. 



In the first place, much time was spent in the search for some 

 possible error in our results. For example, several of our samples of 

 salt, especially the preparation numbered K, were tested for lithium 

 with the greatest care by fractional treatment with alcohol. The 

 extreme mother liquors showed no trace of the lighter metal, although 

 in parallel experiments it was found that less than 0.01 per cent of 

 lithium chloride could be detected in this way, even by the simple 

 flame test. This proportion could not affect the atomic weight of 

 sodium by more than 0.002 ; but since no trace of lithium was found 

 in any of the specimens of our salt tested, it is clear that our low. value 

 of the atomic weight could not be due to lithium. The constant 

 results found with differently prepared samples from different sources 

 seemed to exclude the possibility of the presence of the very few other 

 metals which would yield too low a value for the atomic weight, 

 namely, beryllium, magnesium, aluminium, and calcium. Acid could 

 not have been present, since the material was usually prepared from 

 salt finally crystallized from pure neutral solution, and was always 

 fused in air. It gave, moreover, a wholly neutral reaction to methyl 

 orange. 



Because no flaw could be found in the operations leading to the 

 new results, the conclusion was inevitable that the old ones were in 

 error. In order to prove this without question, experiments were 

 instituted in which each one in succession of the gravest faults already 

 pointed out in Stas's work were imitated, in order to detect the indi- 

 vidual errors introduced by each. 



In the first place, two experiments were made in which the salt was 

 prepared approximately according to Stas's method, the substance 

 being fused with ammonic chlorplatinate. The salt thus prepared was 

 compared with pure silver by means of the nephelometer, using, as 

 Stas did, the opalescence which had been given only fifteen minutes 

 to form. In this way 2.67137 and 5.51270 grams of salt needed, 

 respectively, 4.92908 and 10.1717 grams of silver, corresponding to an 

 atomic weight of sodium of 23.039 in each case. Since the value 

 obtained from salt certainly free from platinum in the same way (p. 45) 



