In Fig. 4 _M_ is a magnet bent into the form of a U, commonly called a horseshoe magnet. The short bar set between the upper ends is also a magnet, and is arranged so as to revolve around the shaft _s_. From what has just been explained in connection with Figs. 1 and 2 it will be understood that, with the poles as indicated by the letters, there will be an attractive force set up between the top end of the straight bar and the _P_ end of the horseshoe, and thus rotation will be produced in the direction of the arrow. The rotation, however, will necessarily stop when the bar reaches the position shown in Fig. 5, for then the attraction between the poles will resist further movement. If the straight bar were not a magnet, but simply a piece of iron or steel, it is evident that when in the position of Fig. 4 the attraction would be just as much toward the right as toward the left, and if the bar were placed accurately in the central position it would not swing in either direction. It would be in the condition called, in mechanics, unstable equilibrium. In practice this condition could not be very well realized, as it would be difficult to set and retain the bar in a position where the attraction from both sides would be the same, therefore the rotation would be in one direction or the other; but whichever way the bar might move, it would only swing through one quarter of a revolution, into the horizontal position of Fig. 5. If we reflect upon these actions we can see that if we could destroy the magnetism of both parts before the straight bar reaches the position of Fig. 5 it would be possible to obtain rotation through a greater distance than one quarter of a turn, for then the headway acquired by the rotating part would cause it to continue its motion. If, after the completion of one half of a revolution, we could remagnetize both parts, we would then set up an attraction between the lower end of the straight bar and the left side of the horseshoe, for then the polarity of the former would be the reverse of that shown in Fig. 4--that is, the lower end would be negative. By means of this second attraction we would cause the bar to rotate through the third quarter of the revolution, and if, just before completing this last quarter, we were to remove all the magnetism again, the headway would keep up the motion through the final quarter of the revolution, thus completing one full turn. From this it will be realized that if we could magnetize and demagnetize the two parts twice in each revolution a continuous rotation could be obtained.