1890 - 91 .] Prof. C. G. Knott on Electric Resistance of Cobalt. 305 
Since the second differences have appreciably different values, it 
is impossible to represent the law of change by means of a parabolic 
function. But the remarkable constancy of the ratios of successive 
pairs of resistances suggests an exponential function of the tem- 
perature as the expression for the resistance. 
Thus we may put, 
r = ae u 
from which we find, if t is the temperature in degrees centigrade, 
k — '002605 a= -09511 
According to this formula, which strictly applies only to tempera- 
tures above 100°, the resistance at 7°*5 should be *09698, almost 
exactly 1 per cent, too high. 
In the paper already referred to, I found that the same form 
of expression held very approximately for the case of one of the 
nickel wires, the only essential difference being in the value 
of k, which for nickel was *003. The resistance of cobalt there- 
fore does not change so quickly with temperature as does the 
resistance of nickel. 
In the second series of experiments, the lower ends of the rods, 
with their connecting wires, were inserted into a porcelain vessel. 
Asbestos was wrapped round the wires ; and the whole was heated 
in a charcoal furnace. The observations of resistance were made 
as the system was cooling, the cobalt and platinum being thrown 
alternately into the Wheatstone Bridge. The instants at which 
the balancings were effected were carefully noted, so that it was 
an easy matter to interpolate between two successive measurements 
for the one wire that resistance which corresponded to the inter- 
mediate measurement for the other wire. In this way, for every 
measured cobalt resistance, the corresponding resistance of platinum 
was calculated by a simple interpolation. After all corrections 
were applied, every resistance was divided by the resistance of 
the same wire at 7° C. By this treatment the results of the 
four different experiments were reduced to identically the same 
condition, so that direct comparison was possible. 
Each single experiment contained from 20 to 30 distinct pairs 
of measurements. These numbers were then classified into groups, 
