640 PROCEEDINGS OF THE AMERICAN ACADEMY. 



found between two points separated by the width of the main plate plus 

 the length of both arms, it might seem as if iv must be taken as this 

 whole distance. On the other hand, as the flow of heat in the arms 

 follows a somewhat dubious course, one may well doubt whether the 

 whole length of the arms should be included. Fortunately, the fact 



rff) 



that the width w occurs in combination with the gradient -^ relieves 



us from this difficulty. The longitudinal gradient which counts is the 

 gradient on the narrow strip running across from arm to arm and out 

 into the arms, and this gradient will bear the same ratio to the meas- 



ured gradient -rr that the width of the main part of the plate bears to 



the effective width at the arms. Accordingly, we can take ir as the 



width of the plate without the arms and -rr as the gradient which 



would exist if the arms were absent, that is, practically, the gradient 

 as measured by means of the circuit Ci -plate- Cs. This we have done, 

 here and in dealing with the Leduc effect. 

 The temperature-coefficient oi ^Tg we find to be 



(980 - 860) -^ 860 (60 - 31) = 0.0048. 



The Leduc Effect. — All of the Leduc observations on Plate 1 were 

 made with the iron trough in use. For this reason, and the further 

 reason that the dubious thermo-electric quality of our iron wire was 

 involved in them, these observations are entitled to far less weight 

 than those made with Plate 2. 



The method used with both plates was essentially that indicated in 

 Figure 9, the (2) circuit and the (4) circuit being used differentially, one 

 connected with the west coil of the galvanometer, the other connected 

 with the east coil. 



With Plate 1 we found 



