FORCES ACTING IN MAGNETIC CIRCUITS. 203: 
(iv.) The state of the surfaceof the barsis an important matter. 
If the ends are rough of course the contact is at the 
points. This leads to a concentration of induction at 
these places, and is a very constant source of error. My 
best bars however gave just as anomalous results as my 
roughest ones, so that the deviation could not be attri- 
buted to this cause. | 
(v.) Finally I tried—without much confidence—the effect of a 
small indirectness of pull. This was done by winding 
two wide solenoids and leaving sufficient space between 
their ends to see what was going on at the plane of 
junction of the iron, a gas flame was put on the side of 
the junction remote from the observer. 
It was then found that one side of the bar which was being 
pulled off invariably remained in contact after the other side had 
slightly separated—when this was prevented by slightly guiding 
the spring balance by hand the agreement was as good as at higher 
inductions. The explanation is now obvious, if the bars separate 
slightly at one side, two things happen: (1) The total reluctance 
of the circuit increases; (2) the induction concentrates at the 
parts in contact. At low inductions the effect of (2) overpowers 
that of (1); at high inductions when the permeability of the iron 
becomes less the induction is less free to distribute itself and also 
the traction of the bars being greater, the phenomenon does not 
begin to manifest itself till rupture is just about to be produced 
or it produces a very much smaller percentage error. Of course 
all this might have been foreseen, but one’s experience with strong 
magnets—in which case it is notorious that it is much more 
difficult to pull off an armature straight than slightly sideways— 
misled me. 
On examining other bars which had given similar results at low 
inductions, I found I could similarly diminish or increase the 
apparent traction by varying very slightly the direction of pull. 
Thus with a pair of flat ended bars and an induction density of 
about 3,000C.G.8. the calculated pull was one pound seven ounces, 
