CONTINUOUS ELECTRIC CALORIMETRY. 
135 
of cooling. At the time of reading the maximum, the zero depression has not had 
time to produce its full effect, owing to the suddenness of the rise, and it therefore 
tends to increase slightly the subsequent rate of fall. In graduating the thermo¬ 
meter, or comparing it with an air-thermometer, the readings are taken at steady 
temperatures, so that the zero depression has time to produce its full effect. These 
steady readings will be systematically lower than the instantaneous readings obtained 
on suddenly heating the thermometer. Unless the method of variable zero is 
employed in taking the observations, it is quite impossible to apply an accurate 
correction, since the depression in any given case depends so much on the past 
treatment of the thermometer, especially in the case of French “ cristal ” glass. It 
is possible, however, to assert that the probable effect would be to produce an error 
in the observed rise of temperature approximately proportional to the rise, and 
therefore nearly proportional to the excess of temperature of the hot water, since 
the same weight of water was employed in all the experiments. The corresponding 
error in the value of the mean specific heat deduced would therefore be nearly 
constant on the average, although no doubt the variations of the zero depressions in 
consecutive experiments may be responsible for some of the individual discrepancies. 
The possible limit of error from this source would be about 5 parts in 1000, allowing 
for the effect of the zero depression in accelerating the apparent rate of cooling, 
which would tend to increase the error. The probable effect would be to make the 
specific heat, as calculated by Regnault, about 2 or 3 parts in 1000 too large. 
The correction of Regnault’s thermometers to the hydrogen scale cannot be 
applied with any certainty without recovering the original instruments, as different 
thermometers of the same glass often differ considerably, and so much depends on 
the exact method of treatment. But if we assume Guillaume’s tables for modern 
thermometers of similar glass, the correction to the values of the specific heat 
would be of -the order of 3 parts in 1000 in the direction of reducing Regn ault’s 
results, and would be nearly constant for the different observations. 
Including both sources of error, we should infer that Regnault’s values for the 
mean specific heat may require to be reduced by a constant correction of 5 or 6 
parts in 1000. 
Although it is evident that some correction is necessary, I should hesitate to 
assume the above estimate without experimental corroboration. 1 find, however, as 
explained in the ‘Brit. Assoc. Rep.,’ 1899, that a correction of precisely this order 
of magnitude is required to make Regnault’s observations of the mean specific heat 
between 20° and 110° C., agree with those of Reynolds and Moorby, between 0° 
and 100°, and with those of Barnes, between 40° and 90° C. The correction also 
makes Regnault’s observations at 110° agree much better with his own observations 
at higher temperatures. It has the further advantage of being the simplest, as well 
as the most probable kind of correction to apply. I therefore proposed in 1899 to 
adopt Regnault’s formula provisionally for the higher temperatures, merely 
