Water between the Freezing and Boiling Points, &c. 243 



ing all values of the cubical term positive. Above 55 the simple 

 formula 



S = 0-99850 + 0-000120 (-55) + 0-00000025 (t- 55) 2 



holds with great accuracy. 



Both these formulae are given for comparison in the second table. 

 They fit very closely except below 5, where the specific heat curve 

 increases more rapidly. These loAver values are within 1 part in 1000, 

 however. 



A summary of the values obtained for the radiation loss shows that 

 the absolute value for any one calorimeter cannot be relied on to an 

 order of accuracy greater than 1 part in 1000 over extended periods. 

 This is particularly true when the temperature of the calorimeter is 

 widely changed. It was found most essential to always eliminate the 

 heat loss from at least two different flows in order to be completely 

 independent of its absolute value. The complete independence of the 

 results from the value of the heat loss, provided this remained con- 

 stant throughout the time of an experiment, was shown by employing 

 calorimeters with different degrees of vacuum involving widely differ- 

 ing values of the heat loss. The temperature coefficient of the radia- 

 tion loss was found to be almost exactly linear over the range of these 

 -experiments. 



Relation to the Work of other Observers. 



It is at once apparent that the value of the mean mechanical equiva- 

 lent of heat obtained from these meastirements, which is 4-18876 joules, 

 is somewhat larger than the exceedingly accurate and trustworthy 

 measurements of Eeynolds and Moorby. Their value, which is 

 4-18320 joules, is lower by 0-132 per cent., or a little over 1 part in 

 1000. 



It is evident that this error may be attributed to the neglecting of 

 some correction factor in the present series of experiments at the 

 extremities of the range, which would cause the variation curve to 

 increase more rapidly than it truly does ; but from the order of accu- 

 racy with which the theory of the present experiments holds at the 

 extremities of the range, it is far more likely that the variation 

 curve is correct, and that the difference in the two results is to be 

 attributed to an error in one of the constants. The thermal constants 

 employed in the two different experiments are referred to the same 

 values, but the introduction of the value of the electrical units into 

 the present series of experiments, which do not enter into the calcula- 

 tion of Eeynolds's and Moorby's result, renders it highly probable 

 that the error is to be looked for here. In view of the immense 

 amount of labour expended in establishing the value of the inter- 



T 2 



