404 PROCEEDINGS OF THE AMERICAN ACADEMY. 



this efifect was as follows. Pressure was pushed to a value somewhat 

 short of the freezing pressure, the readings were made, and then pres- 

 sure increased again, etc., until the subcooled region had been entered 

 far enough for freezing to begin. This first stage might occupy an 

 hour. The second stage was that of approximately constant pressure 

 during freezing, which might require from an hour to an hour and a 

 half During the third stage, pressure was increased a few hundred 

 atmospheres beyond the freezing pressure, held at the maximum for a 

 while, and then decreased to the melting point again. This occupied 

 another hour. The fourth stage, melting at constant pressure, ran 

 more rapidly than freezing, but might consume an hour. Finally a few 

 readings below the freezing pressure, rarely more than half an hour. 

 It is to be noticed that during melting the pressure has been held in 

 the immediate neighborhood of the maximum for some hours, and that 

 this melting pressure is only slightly removed from the maximum, 

 whereas during freezing, the stress has j ust been increased over nearly 

 the entire range and the time for accommodation has been much 

 shorter. All our experience with elastic after-effects would lead us to 

 predict that during melting these effects would be very much less im- 

 portant than during freezing. We would furthermore predict that the 

 change of volume found during freezing would be greater than that 

 during melting, for the piston must advance to compensate for the 

 increasing volume of the steel vessels as well as for the decreased vol- 

 ume of the mercury. 



These conclusions were verified by the data. The values found 

 during melting were always self-consistent, lying on a smooth curve, 

 and could be repeated. But those found during freezing might vary 

 wildly within the 3 per cent limit mentioned above. Furthermore, the 

 inconsistent results were always larger than the consistent ones, pro- 

 vided that the procedure was as described above. There seemed to 

 be no connection between the duration of the pressure and the amount 

 of the discrepancy. 



We already have evidence enough to make it extremely probable 

 that the discrepancy was not due to leak, although it is of the right 

 sign. But the fact that the discrepancy is reversed in sign so as to in- 

 dicate a leak in instead of a leak out, if the stress is applied differently 

 from above, renders certain the conclusion that the effect is not due to 

 a leak. The succession of points on the melting curve, the procedure 

 in getting one of which was described above, was in general obtained 

 by starting at low temperatures and pressures and working up to the 

 higher values. Each maximum pressure, then, was higher than the 

 preceding maximum. But if the preceding maximum was higher than 



