to Calcium and Magnesium Metasilicates. 5 



neither is the best that could be given. Each represents three 

 different charges. 



A word as to the exact significance of the accuracy here indi- 

 cated may avoid some confusion. 



In comparative measurements with very small crucibles (2*5 

 grams) and bare thermoelements, the relative accuracy obtain- 

 able by the same observer with the same apparatus is J° or 

 better ; with the larger crucibles (25 grams) and jacketed ele- 

 ments, differences may reach 3°. This is shown in Table I, 

 page 1. 



The absolute value of the measurements is not established 

 with this accuracy. Different observers, indeed, working inde- 

 pendently with different apparatus, ought not to differ much 

 in their results through differences in experimental methods, 

 provided these are not positively faulty ; that is, provided the 

 thermal junction is actually at the temperature of the melting 

 material. This is shown by the agreement of our earlier and 

 later results, obtained with highly differing methods, and also, 

 perhaps more conclusively, by the agreement found between 

 Dr. Day's special form of element* and the ordinary jacketed 

 element. Differences between independent observers are, how- 

 ever, to be expected in the thermoelement calibration, partic- 

 ularly at the present time, when a long extrapolation is 

 necessary. Practically all temperature measurements above 

 the melting point of copper (1084° C, Reichsanstalt Scale) are 

 now obtained by extrapolating some simple function which 

 has been experimentally established for temperatures below 

 that point. With the same function and different elements, 

 temperatures usually do not vary more than 2° C. at 1,500°. 

 With different functions, the extrapolated curves may diverge 

 as much as 30° C. at 1,500°. An error of 0'5° in the determi- 

 nation of the copper point itself may cause an error of 4° in 

 the extrapolated curve at 1,500°. All these are differences in 

 the interpretation of the experimental measurements, which 

 will probably continue to cause considerable differences 

 between the results of different observers until the gas ther- 

 mometer scale is extended to that temperature with sufficient 

 accuracy, after which they can be readily recomputed to the 

 established scale. The temperatures here given are com- 

 puted from the Reichsanstalt Scale. In brief then, the accuracy 



* Carnegie Institution Publication No. 31, p. 25. Our value for magne- 

 sium silicate is 3° higher than that found by Allen, Wright and Clement, 

 but this difference is only a little greater than the accidental variations 

 occurring in the present work. With charges of the same size, errors in the 

 method seem most likely to occur, if at all, as the result of a very uneven 

 temperature distribution in the furnace or of insufficient immersion of the 

 thermoelement in the charge. In a few cases where the platinum jacket 

 dropped down so as to rest against the bottom of the crucible, which was 

 cooled by contact with the pedestal below it, the melting point came about 

 5° too low. 



