( 820 ) 
In our case the deviation from this is by no means a priori negligible: 
the two limiting values of P(rp=0° and <p = 90") differ in our 
case by 0,3%, and since 1 3 always changes between these two 
limits in the same direction the error caused thereby when 
sin </ cos — % is less than 0.1 %. 
In contrast with the two foregoing sources of error, the reaction 
of the magnetized ellipsoid upon the distribution of magnetism over 
the surface of the pole-pieces can clearly have no effect in the case 
of a body of small susceptibility while on the other hand, it had to 
be taken into account in the casp of the ferromagnetic bodies. Indeed, 
with oxygen we have to deal with a magnetization that in the 
strongest fields of the electromagnet reaches a value of only a few 
units (in the case of iron it was 1700!). 
3. Influence of the holder. In this connection we may notice two 
actions that may go together. In the first place there is the inherent 
magnetism of the stem, and then there is also an action analogous 
to that which we wish to measure, for if the stem is not a perfect 
body of revolution, it is acted upon in the liquid oxygen just as if 
it were a supplementary ellipsoid. We investigated these two sources 
of error in a blank experiment in liquid oxygen in which the silver 
ellipsoid was removed, and the surface of the glass was carefully 
freed from all traces of wax. From this we obtained a maximum 
of only 1 to 2 %o which need not be taken into account 
4. The concentration of the oxygen. The oxygen in the bath contained 
a little nitrogen, the concentration of which constantly decreased during 
the experiment owing to its faster vaporization. So as to be able to 
allow for this we analysed the gas at the beginning and at the end 
of each series of observations. The mean concentration was 1.25% 
at the beginning and 0.35% at the end (at the moment that the Dewar 
vessel was almost empty). We allowed for this concentration as far as 
possible; in this respect there remains an uncertainty of about 0.3%. 
5. Calibration of the suspension springs. The main torsion spring 
described above was calibrated outside the apparatus by observing 
the time of oscillation of a system suspended from it with and 
without the addition of a known moment of inertia. For the latter 
we used a bronze ring of rectangular meridian cross section, the 
diameters and height of which were measured with the cathetometer. 
Calculation gave the moment of inertia as 
582.09 c.g.s. 
Care was taken that the spring was subjected to the same tension 
during the calibration as it experienced while in the apparatus (by 
attaching suitable weights to it by a torsion-less wire). 
