[cookr] VARIATION IN THE DENSITY OF ICE 138 
certain thickness, as the formation invariably became filled with a 
network of fine cracks when the mantle became very large. The most 
probable explanation of this phenomenon would seem to be that when 
the glass support and the inner layers have practically reached the 
temperature of the refrigerant, the outer layers are still at a tem- 
 perature very near the freezing point. Hence, as the freezing pro- 
gresses, the outer layers tend to contract far more than the inner ones; 
this contraction is free to take place in a radial direction, but the 
circumferential dimensions of the outer layers are fixed when the 
supporting layers have reached the limit of contraction. Therefore 
a decrease in temperature of these superimposed layers should result 
in the formation of small cracks, the general plane of these cracks 
being in the direction of the length of the tube and extending radially 
outwards. With regard to Professor Nichols’ determinations on ice- 
mantles by the method of weighing, it seems highly improbable that 
these mantles were preserved intact during the experiments, and, if so, 
the results can have no bearing on this discussion, since a breaking up 
of the ice-mantle would at once release it from the effect of any inter- 
nal strains. 
Two determinations were also made by Nichols on the density 
of very fine specimens of the icicle formation, the result obtained, 
using the method of weighing, being 0:91807 + -00004. 
Both Brunner and Nichols have determined the density of ice 
taken from the surface of ponds, formed at the beginning of very cold 
weather, and these two determinations show a remarkable agreement. 
Nichols also experimented with pond ice one year old, obtaining a 
value considerably lower than that of the newly cut specimens. 
Bunsen, using an overflow dilatometer and experimenting on the 
column formation, obtained 0:91685 + -00003. 
As the result of seven determinations of the density of the St. 
Lawrence ice, made by Professor H. T. Barnes and the author, the 
value 0:91661 + -000065 was obtained. The ice used in these measure- 
ments was cut from the lower surface of the ice sheet, and the speci- 
mens ranged from a month to over two years in age. No trace of 
any variation due to age was detected. A modification of the method 
of weighing was used. 
These results are all given in tabulated form below. In the 
sixth column will be found the probable result, whether high or low, 
according to the theory given in this paper, and in the seventh column 
the difference from the density of old river ice is given, the value 
assumed being a mean between Nichols value and our own, weights 
