262 



PROCEEDINGS OF THE AMERICAN ACADEMY. 



system absolutely to rest was no small one and yet extremely impor- 

 tant for accurate measurements. 



The data for three curves was obtained with the maximum flux 



densities respectively of 800, 

 7650 and 14,000 per square 

 centimeter. Since the method 

 was almost exactly similar to 

 that employed in the case of 

 toriods I and II, it seems 

 worth while to note only the 

 difference. This lay in the 

 demagnetization ; for toroid 

 III it was carried out wholly 

 by the reversing switch, N 

 (Figure 3), and the rheostat 

 or resistance cells, R and 

 RG. 



The accuracy with which 

 the long-period galvanometer 

 could be handled was fully as 

 great as that of the shorter. 

 The only error introduced lay 

 in the 0.7 of one per cent; 

 and this was eliminated to 

 considerable extent. For 



20 



6 REVERSALS 



Figure 20. The value of B is 7650: the 

 X indicates 22-step reversals, the O 1-step 

 reversals. 



a 



while about 0.7 of one per 

 cent of the throw would be 

 lost when 1-step reversal was 

 taken, a corresponding loss occurred in the reversals consisting of a 

 larger number of steps, for the same field reversed. It was found that 

 a large proportion of these 

 small steps required about 

 20 seconds for their com- 

 pletion ; as a conseijuence 

 their sum would be less 

 than the true sum. From 

 observations as to the time 

 the two instances and 



240 



in 



239 



1 REVERSALS 5 X 



Figure 21. The value of B is 14,000: the 

 X indicates 28-step reversals, the O 1-step 

 reversals. 



the size of the throws I 



should regard it as very 



probable that at least half 



the error was cancelled by this introduction of the same inaccuracy in 



the two cases compared. 



