94 BRIDGMAN. 



a thermal conductivity at least fifteen times less. The error from 

 heat leak along the wires of the couple should therefore be very much 

 less with the therlo couple. As a matter of fact, the same results 

 were obtained w'ith both couples, establishing freedom from heat leak. 



A central band of one or two coats of enamel (thickness not over 

 0.0001 inch) was usually baked on the outside of the specimen. This 

 was to avoid danger of short circuit where the thermo-couple wires 

 were wrapped about the cylinder. Baking on this band of enamel, 

 which required about 210° C, served the additional purpose of anneal- 

 ing the specimen. All the specimens were further heated to 150° or 

 so when soldered into the massive copper block. 



As already explained, the dimensions were forced by compromise 

 between various opposing tendencies. If it were not for lateral loss, 

 the specimens would have been made much longer, and the source and 

 sink would have been situated further from the couple. This would 

 have allowed the lines of flow to more completely straighten out in the 

 vicinity of the couple. It is possible by a rough calculation to get an 

 approximate idea of the extent of the failure of the lines of flow to be 

 entirely straight. The actual distribution of temperature over the 

 section at the source end cannot of course be accurately determined, 

 but it seems fair to assume that the temperature is higher at the center 

 of the rod than at the outside surface. A solution corresponding to 

 this state of affairs may be obtained by assuming a distribution of 

 temperature according to a Bessel's function over the source end. I 

 carried through an approximate examination in this way, and con- 

 vinced myself that any errors from this eflFect were not important. 



The largest correction to be applied in the longitudinal flow method 

 is for the lateral loss of heat through the transmitting medium. The 

 lateral loss is a small part of the total input, so that we may assume 

 the distribution of temperature approximately linear along the speci- 

 men. The temperature gradient at the mean point between the two 

 junctions is the dift'erence of temperature divided by the distance 

 between the junctions. The total flow of heat long the rod at this 

 point under this gradient is not the total input, because there has been 

 some lateral loss between the source and the junction. This loss takes 

 place partly from the lateral curved surface, and partly from the end. 

 The loss from the lateral curved surface of the bar was competed on 

 the basis of the formulas for the radial flow of heat through a cylinder 

 between inner and outer surfaces maintained at constant difference of 

 temperature. The temperature of the inner surface, that is, the 

 temperature of the bar, varies along the axis. The total lateral loss 



