96 A. L. Day and R. B. Sosman — 



to stress. If the bar was cooled down very slowly, it returned 

 nearly to its initial length ; if cooled rapidly, it required sev- 

 eral days to return to its original dimensions. This irregularity 

 makes up most of the 0*5 per cent uncertainty mentioned 

 above. 



Plan. — Above 1100° considerable uncertainty has existed 

 in the temperatures of various fixed points. The melting 

 point of nickel, considered as 1484°,"* has been frequently 

 employed. The carve of the platinum-rhodium thermoele- 

 ment, extrapolated beyond the copper-point, has been still 

 more generally used, but like any extrapolation, may lead to 

 quite erroneous results. The only gas thermometer compari- 

 son that has been made in this region is that of Holborn and 

 Valentiner,f but by their own estimate the accuracy of the 

 upper portion of their scale is not greater than ± 10°. The 

 chief purpose of our work was, therefore, to establish the tem- 

 perature of several fixed points between 1100° and 1600° and 

 to find what curve is followed by the platinum-rhodium ther- 

 moelement in this region, with an accuracy comparable to that 

 obtained in the lower portion. 



The plan of the work is simple. It consists, first, in select- 

 ing certain fixed therm ometric points, usually melting points 

 of metals, and in determining their reproducibility ; second, 

 in making a measurement of the true temperature on the 

 nitrogen scale at or close by one of these fixed points ; third, 

 in transferring this known temperature by means of a thermo- 

 element over to the fixed point in question. This transference 

 by the thermoelement is necessary because the thermometer 

 bulb cannot be put directly into melting or solidifying sub- 

 stances at high temperatures. The relation of electromotive 

 force to temperature for any particular kind of thermoelement 

 does not enter into the problem when the temperatures 

 measured are close to the fixed points; a linear correction is 

 then abundantly accurate. The interpolation curve, for any 

 element, between the fixed points established by the gas ther- 

 mometer, is therefore a separate matter. 



The questions which remain to be answered are, then : (1) 

 How exact and uniform can the temperature of the gas in the 

 bulb be made (independently of any effort to measure this 

 temperature) % (2) How accurately can its pressure be meas- 

 ured in order to establish that temperature on the nitrogen 

 scale ? (3) How accurately can this temperature be transferred 

 from the thermometer and compared with the fixed melting 

 point ? (4) How accurately can the fixed points be reproduced 

 for purposes of calibration of secondary measuring devices ? 



As has been stated, our experience has convinced us that 

 the most of the variations in the gas scale temperatures of the 



* Holborn and Wien, Wied. Ann., xlvii, 107, 1892 ; and lvi, 360, 1895. 

 f Ann. d. Phys. (4), xxii, 1, 1907. 



