OF ARTS AND SCIENCES. 171 



The variation which I find is almost exactly that given by McFar- 

 lane, as is shown by the constancy of the column of ratios. Hut my 

 coefficients are less than half those of McFarlane. This may possibly 

 be due to the fact that the walls of McFarlane's enclosure were 

 blackened, ami to his surface being of polished copper and mine of 

 polished nickel : his surface may also have been better adapted by its 

 form to the loss of heat by convection. The results of Nichol are 

 also much lower than those of McFarlane. 



The fact that the coefficients of radiation are less with increased 

 temperature of jacket is just contrary to what Dulong and Petit found 

 for radiation. But as I have shown that convection is the principal 

 factor, I am at a loss to check my result with any other observer. 

 Dulong and Petit make the loss from convection dependent only upon 

 the difference of temperature, and approximately upon the square root 

 of the pressure of the gas. Theoretically it would seem that the loss 

 should be less as the mean temperature rises, seeing that the air be- 

 comes less dense and its viscosity increases. Should we substitute 

 density for pressure in Dulong's law, we should have the loss by con- 

 vection inversely as the square root of the mean absolute tempera- 

 ture, or approximately the absolute temperature of the jacket. This 

 would give a decrease of one per cent in the radiation for about 6°, 

 which is not far from what I have found. 



To estimate the accuracy with which the radiation has been obtained 

 is a very difficult matter, for the circumstances in the experiment are 

 not the same as when the radiation was obtained. In the first place, 

 although the water is stirred during the radiation, yet it is not stirred 

 so violently as during the experiment. Further, the wheel above 

 the calorimeter is warmer during radiation than during the experi- 

 ment. Both these sources of error tend to give too small coefficients 

 of radiation, and this is confirmed by looking over the final tables. 

 But I have not felt at liberty to make any corrections based on the 

 final results, as that would destroy the independence of the observa- 

 tions. But we are able thus to get the limits of the error produced. 



During the preliminary experiments a water jacket was not used, 

 but only a tin case, whose temperature was noted by a thermometer 

 above and below. The radiation under these circumstances was 

 larger, as the case was not entirely closed at the bottom, and so per- 

 mitted more circulation of air. 



3. Corrections to Thermometers, etc 

 Among the other corrections to the temperature as read off from 

 the thermometers, the correction for the stem at the temperature of 



