1903 - 4 .] Thermal Expansion of Solutions of Hydroxides. 285 
indication that the one method is any better than the other from 
my point of view. I therefore did not use the counterpoise. 
In the determinations of the densities of the solutions, the 
pyknometers weighed about 20 grams, and had a capacity of about 
20 c.c. The pyknometers, after being filled, were placed in a 
thermostat, the temperature of which was kept at 15° C., 20° C., 
26° C., 30° C., as was required ; the bath did not vary more than 
'04° C. from the required temperature during any experiment. The 
stirrer was driven by an electric motor, or latterly by a Heinrici 
hot-air engine. The thermometer which gave the temperature of 
the hath was graduated to fiftieths of a degree centigrade, and had 
a table of corrections from the National Physical Laboratory, New 
Observatory. After the pyknometer had been for some time in 
the bath (the period varying from 2 hours to 20 hours, as the 
apparatus was kept going day and night), the meniscus was made 
to coincide with the mark on the stem. A short time after, if the 
meniscus still coincided with the mark, the pyknometer was taken 
out, dried with a cloth and weighed. All weighings were corrected 
for the buoyancy of the air by adding on to the observed weight 
of the pyknometer the weight of air displaced by the excess of the 
volume of the pyknometer and liquid over that of the weights. 
To get an accuracy of ‘001 per cent, in a weighing the 
thermometer in the balance-case should be read to T4° C. and 
the barometer to *35 mm The thermometer in the balance-case 
read to T° C. and was correct to ’02° C., and the air in the case 
was kept dry by means of sulphuric acid. The barometer, which 
had been corrected at the National Physical Laboratory, read to 
T mm. In the correction for buoyancy the density of the air 
was taken from Landolt and Bornstein.* The error introduced 
by taking the air in the balance-case as perfectly dry was 
calculated and found to be negligible. 
All weighings were the means of at least two observations, and 
the deviation of any weighing from the mean of two weighings 
was found not to exceed ‘002 per cent, for 94 weighings examined, 
thus giving a rough estimate of the accuracy in weighing. 
The so-called “ probable error ” in the estimations of density 
was found not to exceed ‘00002. 
Loc. cit. 
