178 PROCEEDINGS OF THE AMERICAN ACADEMY 



degrees. The tables so formed I have combined in two ways : first, 

 I have added to the column of work in each table an arbitrary number, 

 such as to make the work at 20° about 10,000, and have then combined 

 them as seen in Table LI. ; and, secondly, I have subtracted each 

 number from the one 1 0° farther down the table, and divided the num- 

 bers so found by 10, thus obtaining the mechanical equivalent of heat. 



In these tables four thermometers have been used, and yet they 

 were so accurate that little difference can be observed in the experi- 

 ments which can be traced to an error of the thermometer, although 

 the Kew standard has some local irregularities. The greatest difference 

 between any column of Table LI. and the general mean is only 10 

 kilogramme-meters, or 0.023 degree, and this includes all errors of 

 calibration of thermometers, radiation, &c. This seems to me to be a 

 very remarkable result, and demonstrates the surpassing accuracy of 

 the method. Indeed, the limit of accuracy in thermometry is the only 

 limit which we can at present give to this method of experiment. 

 Hence the large proportional time spent on that subject. 



The accuracy of the radiation is demonstrated, to some extent, by 

 the agreement of the results obtained even with different temperatures 

 of the jacket. But on close observation it seems apparent that the 

 coefficients of radiation should be further increased as there is a ten- 

 dency of the end figures in each series to become too high. This is 

 exactly what we should suppose, as we have seen that nearly all 

 sources of error tend in the direction of making the radiation too 

 small. For instance, an error came from not stirring the water dur- 

 ing the radiation, and there must be a small residual error from not 

 stirring so fast during radiation as during the experiment. Besides 

 this, some parts around the calorimeter were warm during the radiation 

 which were cool during the experiment. And both of these make the 

 correction for radiation too small. However, the error from this 

 source is small, and cannot possibly affect the general conclusions. In 

 each column of Tables LI. and LI I. a dash is placed at the tem- 

 perature of the jacket, and for fifteen degrees below this point the 

 error in the radiation must produce only an inappreciable error 

 in the equivalent: taking the observations within this limit as the 

 standards, and rejecting the others, we should still arrive at very 

 nearly the same conclusions as if we accepted the whole. 



Most of the experiments are made with a weight of about 7.3 klU as 

 everything seemed to work best with this weight. But for the sake 

 of a test I have run the weight up to 8.G and down to 4.4 ki1, l>y which 

 the rate of generation of the heat was changed nearly three times. 



