172 HENKY A. EOWLAND 



measurement from the end of the bar, R, to the sides of the circle, F. 

 The adjustment was always within 30', which would only cause an error 

 of one part in 25000. 



The needle must hang in the magnetic meridian by a fibre without 

 torsion, and the coils must be parallel to it. These adjustments were 

 carefully made, but, as has been shown, the error from this source is 

 compensated. 



The needle must hang in the centre of the galvanometer coils and 

 on the axis of the circle. The error from this source is vanishingly 

 small. 



The scale must be perpendicular to the line joining the zero point 

 and the galvanometer needle, it must be level and not too much below 

 the galvanometer needle. All errors from this source are partially or 

 entirely compensated by the method of experiment. 



The induction coils, L, must be horizontal, and at the same level as 

 the two galvanometers, so as not to produce any magnetic action on 

 them. The error from this source is exactly compensated by this 

 method of experiment, but could never amount to more than 1 part in 

 2000. 



The tangent galvanometer should have the plane of its coils in the 

 magnetic meridian, but all errors are compensated. 



The connecting wires must be so twisted together and arranged as 

 to produce no magnetic action, but tests were made in all cases where 

 the error was not compensated, and found to be practically zero. The 

 insulation of all coils, wires and commutators was carefully tested. 



Method of experiment. As has been stated before, the method gener- 

 ally used was that of the first throw of the needle, though the method 

 of recoil was also used. For the successful use of the first method a 

 quickly vibrating needle and the damping inductor are indispensable, 

 seeing that with a slow moving needle we can never be certain of its 

 being at rest. By this method it is not necessary to have the needle 

 at rest at the zero point, but, if it vibrates in an arc of only a millimetre 

 or two, we have only to wait till it comes to rest at its point of greatest 

 elongation on either side of the zero point and then reverse the commu- 

 tator. The error by this method is in the direction of making the 

 throw greater in proportion of the cosine of the phase to unity. The 

 smallest throw used was 100 mm. Hence, if the needle vibrated 

 through a total arc of 2 mm., the error would be 1 in 17,000. In reality 

 the needle was always brought to rest much more nearly than this. 



The method of recoil was used once with the needle vibrating in 7-8 



