618 PROCEEDINGS OF THE AMERICAN ACADEMY. 



surrounding the bars, consequent upon a reversal of the gradient of 

 temperature in the apparatus. 



Already, in connection with the discussion of loss of heat by lateral 

 outflow, the question has been raised whether this loss is merely, or 

 practically, a function of lateral differences of temperature. If we had 

 to do here with a true conduction only, there would be little risk in 

 answering this question in the affirmative ; but when convection, accom- 

 panying rather general movements of the air within the packing, comes 

 into play, as it probably does to some extent, the problem becomes more 

 difficult. But it is not easy to find any plausible reason why this 

 longitudinal temperature gradient should be less eff"ective in carrying 

 off" heat from the warmer bar in condition E than in condition W. It 

 is true, that a leakage of naphthalin vapor into the packing space while 

 condition E prevailed might reduce the effectiveness of the convection 

 movements by replacing air by a more sluggish medium ; but in the 

 one case in which such leakage was known to occur, the value obtained 

 for ve was smaller than usual, indicating, if anything concerning lateral 

 loss, that such loss was greater on this occasion than usual. The gen- 

 eral matter of lateral loss will be discussed once more later in this paper. 

 It is the most troublesome part of our experimental problem, and it 

 may be that some inequality of packing of the asbestos fibre is at the 

 bottom of the discrepancy which we are now discussing. 



5. Change of quality from one end to the other of the main bars. 



Both bars were made fi-om pieces taken from the interior of a very 

 soft bar some 10 cm. in diameter, and any diff"erence of composition or 

 physical state sufficient to make such a diff"erence in the Thomson 

 eff"ect with reversal of temperature gradient is very improbable. It is, 

 perhaps, worth noting that in our work of the year 1905 with these 

 same bars, and with the same spirals upon them, what corresponds to 

 our Vg of 1906 was, apparently, about two per cent greater than what 

 corresponds to our v^ of 1906; that is, in both 1905 and 1906 the 

 data obtained with the marked ends (bar a is stamped with a 1 and 

 bar ^ with a ^ on one end) of the bars the hotter have indicated a 

 slightly greater value of the Thomson eff"ect. This is probably an acci- 

 dental coincidence. 



This long search for an explanation of the discrepancy in question, 

 though it has not been decisive, has served the useful purpose of 

 passing in review a number of possible sources of error. A difference 

 of five per cent in the values of i'k and iv is of no great consequence, 

 if we can feel reasonably sure that the mean of the two is not more 

 than two or three per cent different from the true value of v. 



