HOLMAN, LAWRENCE, BARR. — MELTING POINTS. 231 



teuth of one per cent iu the measurement of 2e above 200°C. Tests, 

 check measurements, and a discussion of the sources of error, unne- 

 cessary to detail here, have given satisfactory demonstration that an 

 even higher accuracy than this was attained. As far, therefore, as 

 constant or variable instrumental errors are concerned, it is believed 

 that no error beyond 0°.5 to 1° C. exists in the results, while probably 

 this estimate is large. 



The error from impurities must have been exceptionally small, as 

 the analysis of the metals indicates. Some impurities from alloying 

 with the platinum and rhodium of the thermo-couple must have en- 

 tered during the experimenting, but as results at difFei'ent stages of 

 the work checked those obtained upon the first use of the metal, and 

 as reuewals of the metal made no difference in readings beyond the 

 limits of other variations (about 5 parts in 10,000), the error from 

 this source must have been negligible. 



In the case of platinum the metal at command was unfortunately 

 not of known composition, nor was it possible at the time to obtain 

 any whose purity was known. An analysis of the wire used may 

 perhaps be obtained later, and it is hoped to carry out further meas- 

 urements with the better platinum now obtainable through the recent 

 advances made in its manufacture in Germany and England. 



The aluminum was of very high grade, but it is thought that still 

 better may be obtained, and the peculiar occurrence attending its 

 melting point measurements should be further investigated. 



The actual effect of the small impurities cannot be numerically esti- 

 mated, but must have been inconsiderable except for platinum, where 

 the error probably has the positive sign. 



As to the third point, there was no reasonable doubt left in the 

 minds of the observers that the observed temperatures were sensibly 

 the melting points. Except as noted for aluminum, the readings with 

 rising and falling temperatures did not exceed about one part in one 

 thousand. Also entirely independent observations on separate days, 

 and with renewals of the metals in some cases, were equally concor- 

 dant. The average difference was much less than the error of read- 

 ing the ammeter. As an example of the concordance, and at the 

 same time as showing the homogeneity of the thermo wire, three 

 calibrations in sulphur are quoted in Table III. 



