Permanent Magnets. 145 



much improved, a set was made with each piece separately. 

 The moments had all fallen, but in proportion ; so that there 

 was no evidence of induction interfering in the values which 

 are given. 



The constant of the bifilar was determined by vibrating a 

 leaden rectangle having the same weight as the tray and 

 magnet, and a known moment of inertia. The same suspen- 

 sion-stirrup was used in both cases ; its moment of inertia 

 was determined separately. An approximate value of H was 

 employed. 



The constant of the torsion-balance was determined by 

 vibrating a lead rectangle. The variation of the torsion with 

 the weight was inconsiderable in amount with the wire now 

 employed, but was somewhat irregular. It is very difficult to 

 get wire at all suitable for this purpose. Besides the variation 

 of the constant of torsion, most wires take sets in loading and 

 unloading to such an extent as to make them nearly useless. 

 The best wire I have had is excessively hard. By dint of 

 great labour reliable readings were obtained. 



The first numbers given for 1$, and the figure*, represent 

 the results of experiment. Rowland's method was employed; 

 ten induction-coils for ^=18, 9, 6, 3, 2, and fifteen for the 

 singles. It is clear, however, and was so from a preliminary 

 set, that these results need correction. For, according to the 

 course of the numbers, the induction would vanish, and then 

 become negative for finite lengths, which is inconceivable. 



The source of the error is doubtless in the fact that, with 

 short lengths, the coils cannot be sufficiently concentrated 

 about the equatorial sections. The coils, though bound up 

 closely, extend over a considerable part of the length of the 

 short bars. I have found by direct experiment that spreading 

 the induction-coils leads to considerable diminution of the 

 observed induction. In the present series it was impossible 

 even to fit the 10-coil closely to the bars, as the packing had 

 to pass within the coils. But the coil of 15 was packed so as 

 to fit the singles very closely indeed : the error of the corre- 

 sponding value was thus reduced to about two thirds of its 

 former amount. 



There can be no doubt that the straight line which marks 

 the course of the lower inductions ought to go to the origin, 

 instead of falling just below r it (see figure above). Then up 

 to n = 6 we have t$=w46. This assumes that the value for 

 ?i = 6 is correct, since the error due to the coil must be by this 

 time small. The strengths of the different short bars varied a 



• As originally drawn. The line in the woodcut passes through the 

 origin instead of a very little below it. 



Phil. Mag. S. 5. Vol. 18. No. 111. Aug. 1884. L 



