532 Prof. H. L. Callendar on a 



by my own experiments in 1886, by the experiment? of 

 Griffiths in 1890, and by the recent experiments of Harker 

 and Chappuis at the International Bureau, is far closer than 

 that of any other similar formula in the whole range of 

 thermometry. At higher temperatures, from the B.P. of zinc 

 to the F.P. of copper, the same formula is in agreement with 

 the probable mean of all the best measurements at present 

 available. At still higher temperatures, beyond the present 

 range of the gas-thermometer up to the F.P. of platinum, I 

 have recently succeeded in obtaining presumptive evidence 

 of its validity from a comparison of the scales (1) and (2) 

 furnished by the expansion and calorimetric methods. If we 

 assume that the specific heat of platinum, the coefficient of 

 expansion, and the temperature-coefficient of the resistance, 

 are all linear functions of the temperature, we obtain results 

 which, as shown in Table II., are concordant within the limits 

 of probable error of the measurements. It has been already 

 explained that the results of Violle by method (2) probably 

 require to be raised. The same is probably true of the two 

 highest points by the expansion method, because, owing to 

 the risk of straining the wire, it was necessary to make the 

 tension very light at high temperatures, and the wire was not 

 perfectly straight. 



The differences between the platinum scales (1), (2), (3), 

 and (4) and the scale of the nitrogen-thermometer, are 

 graphically exhibited in fig. 1. These differences are referred 

 to a fundamental interval of 1000° instead of 100°, because 

 the deviation of the scale of the thermocouple at low tempera- 

 tures is so great that it could not conveniently be included in 

 the same diagram, if referred to the usual fundamental interval 

 of 100° C. It will be observed that the difference-curve 

 happens to be the same within the limits of experimental 

 error for methods (1) and (2) ; and that this curve, like that 

 of method (4) , is a symmetrical parabola. The difference- 

 curve of the thermocouple, on the contrary, is not symmetrical, 

 and cannot be satisfactorily represented by a parabolic, or 

 even by a cubic, formula. The favourite type of empirical 

 formula for the thermocouple appears to be either logarithmic 

 or exponential (cf. Becquerel, Barus, Holman, Paschen, 

 Stansfield) ; but the question is still uncertain, and the results 

 of extrapolating different formulas are very discordant, as 

 shown, for instance, by the 1892 and 1894 reductions of 

 Barus's observations in Table II. This appears to be an 

 additional objection to the adoption of the thermocouple as a 

 practical staudard even at high temperatures. 



