386 ROMBERG. 



in which the quantity 5 is as yet unknown. A thermometer cali- 

 brated by the Bureau of Standards at ^Yashington was obtained, its 8 

 being given as 1.47 =t .01. By comparison of the two, the 8 of the 

 thermometer in whose scale these results are given was found to be 

 1.53. This value, when put in the Callendar formula, gives as net 

 corrections in the intervals of 17°-23° and 70°-76°, which were ap- 

 proximately the intervals used in the last series of runs, the values 

 —0.055° and 0.042° respectively, increasing Co— Ci/co from —0.0201 

 to —0.0040, whence the ratio of the specific heats Ci/co is 1.0040. In 

 its relation to the results of previous observers this is shown by a 

 circle on the 73° line of figure 1 . 



The average deviation from the mean, of the last six runs weighted 

 as before, is 0.0007. The "probable error" is about 0.0003. The 

 error arising from the reduction of thermometer ?^ 5 to ^ 3 cannot well 

 be greater than 0.0002. A larger error might have crept in through 

 the 8 correction. If this be due to an error of 4% in 8, the change in 

 Ci/c2 would be a little less than 0.0007. In this connection it is to be 

 noted that the value 1.53 for 8 is within about 4% of the lowest value 

 ever found for platinum, and within 2% of the 5 values commonly 

 found. Since a decrease in 5 would result in raising the point plotted 

 in figure 1, no possible uncertainty in 5 could raise that point above 

 about 1.0047. If the "probable error," namely 0.0003, the estimated 

 limit of error from the thermometer comparison, namely 0.0002, and 

 the error due to a 2% uncertainty in 8, namely about 0.0003, be added, 

 the total uncertainty in Ci/co would be about O.OOOS. The root mean 

 square of these errors would be a little less than 0.0005. 



It will be noted from figure 1 that the author's value, 1.0040, agrees 

 very closely with Dietrici's and fairly closely with that of Barnes, 

 which has been corroborated by Callendar. Bousfield's value, on the 

 other hand, lies close to the older work of Liidin and of Regnault, and 

 the formula of Jager and von Steinwehr, if extrapolated from 50° to 73°, 

 gives 1.0088, which happens to agree almost exactly with Bousfield. 

 These results fall, therefore, quite definitely into two groups, one of 

 which averages about half a percent higher than the other. Now it 

 happens that all four of the methods that led to the high results 

 involved the use of calorimeters containing a stationary pool of water 

 with a free surface,^ while of the four low values, three were obtained 



2 The methods of Janke, Cotty, and BartoU and Stracciati are also of this 

 type, and their results also run high. 



