EFFECT OF PRESSURE OX CONDUCTIVITY OF METALS. 97 



heating current. Both of these checks prove freedom from convective 

 effects. 



The longitudinal flow method showed itself a great improvement 

 over the radial flow method with regard to the agreement of the 

 results obtained with different specimens of the same metal. The 

 individual readings for a single specimen, however, were very much 

 more irregular, and until the interpretation of the irregularities was 

 found, I was in doubt whether this method was actually any improve- 

 ment over the previous one or not. After readings had been made on 

 a number of specimens, it appeared that in almost all cases the read- 

 ings for a single specimen would lie on three or perhaps four straight 

 lines. The extreme separation of these lines amounted to a difference 

 of not more than o% in the total thermal conductivity. It has already 

 been mentioned that this is the variation which might be caused by a 

 shift of position of the thermo-couples in their holes. The natural 

 interpretation is that the couple may take one of two extreme posi- 

 tions, like the bottom of an oil squirt, bent so as to touch either the 

 top or the bottom of the hole. The couple may change from one to 

 the other of these positions with change of pressure. The two couples 

 together can thus occupy any one of three different positions, so that 

 the readings will lie on one of three lines. If the initial curvature of 

 the wires of the couples is such that one of the positions of equilibrium 

 is not exactly at the top or bottom of the hole, then there will be four 

 relative positions of the couples, and the results will lie on four lines, 

 two of them comparatively close together. Figure 9, given later in 

 the detailed discussion of the data for nickel, is a good example of the 

 tendency to lie on discrete lines. 



After the reason for the irregularities had been found it was possible 

 to a considerable extent to control the location of the readings on one 

 or the other of the possible lines. It was found, as might be expected, 

 that rapid changes of pressure were favorable to a jump of the readings 

 from one line to another, but that by changing the pressure very 

 slowly it was in many cases possible to obtain long successions of 

 points on a single straight line. It is easy to see that the nature of the 

 irregularity is such that it was quite possible, before its true nature 

 was recognized, to obtain results which might be erroneous even as 

 to the sign of the effect. Suppose for example that the true effect 

 gives points lying on a falling curve, but that as pressure increases the 

 thermo-couples snap over in the holes in such a way as to pass from 

 the low lying to the high lying curves. If only a few readings are 

 taken, and if the snapping about of the couples is a fairly regular 



