96 BRIDGMAN. 



linear compressibility in the direction of an increase to the apparent, 

 conductivity. In the absence of direct determinations of compressi- 

 bility at high pressures, the corrections applied in the following were 

 then taken from Richard's determinations at low pressures, neglecting 

 the changes of compressibility with pressure. The correction is in 

 any case small; the value for bismuth is a maximum at 1.2%. In 

 addition to these corrections there was an effect not present appreciably 

 with the radial flow method due to loss of heat along the leads to the 

 heating element. The leads were of copper and the heating element 

 was of nichrome. The thermal conductivity of copper is about 30 

 times and the electrical conductivity about 50 times that of nichrome. 

 Hence most of the heat input is confined to the nichrome and there 

 cannot be much heat conduction out along the leads; any cooling of 

 the copper leads can affect only the extreme ends of the nichrome coil. 

 The correction for this effect must be small, but may affect the abso- 

 lute conductivity. Only the change of this correction with pressure 

 can affect the final results, and I have neglected it. 



There is one source of error that might be anticipated to be serious 

 with the longitudinal flow method, namely convection effects in the 

 liquid. In any event the convection effects may be expected to 

 vanish at high pressures because of the known large increase of vis- 

 cosity of the transmitting medium. I did expect the elfect to be 

 appreciable at atmospheric pressure, however, and at first made no 

 reading at atmospheric pressure, but used 2000 kg. as the zero. It 

 later appeared, however, that points obtained at atmospheric pressure 

 lay on a smooth curve with those at higher pressures, so that no error 

 from this effect is to be expected. It is in any event to be remarked 

 that the error from convection may be made vanishingly small by 

 cutting down the heat input. Assuming that the convection loss is 

 only a small part of the conduction loss, we see that if the heat input 

 is halved the temperature differences in the liquid are halved, the rate 

 of convective flow of the liquid is halved, and the rate of heat flow into 

 the liquid is halved, so that the total heat carried away convectively 

 is quartereil. (Of course the loss of heat by conduction to the liquid 

 is only halved). In the actual experiment the magnitude of the heat- 

 ing current was so chosen that the temperature of the source end of the 

 specimen was about 5° higher than ihat of the pressure cylinder. In 

 one or two cases parallel runs under pressure were made with changes 

 in the heat input by a factor of 2, with no change in the results. Also 

 at atmospheric pressure I verified that the temperature difference 

 measured on the thermo-couple was proportional to the square of the 



