BLACKWELL. — DISPERSION IN ELECTRIC DOUBLE REFRACTION. 661 



spectrum taken on successive strips of the same plate. With, every- 

 thing in adjustment, hut with no electric field, the zero band was taken ; 

 with the same adjustments the voltage band was taken, with the 

 electric field; and the reference spectrum was secured by temporarily 

 inserting, immediately in front of the compensator, a sodium flame, 

 and then a cadmium spark. The slit was 0.3 mm. broad, but even with 

 such a broad slit the voltage and zero bands each had to be exposed 

 for at least forty minutes. With this breadth of slit, however, the 

 position of any wave-length could be determined with sufficient accu- 

 racy, as it affects the results only as the square-root approximately. A 

 typical photograph is shown in the Plate, Figure C. During the ex- 

 posure of the voltage band, readings of the potential were taken nearly 

 every minute. Immediately before and after taking a voltage band the 

 static voltmeter was calibrated with the potentiometer and one of three 

 concordant Clark's standard cells. In the worst case the calibration 

 curves differed by four tenths of one per cent, and generally they 

 coincided. 



But the measurement of the negatives proved slower and more 

 troublesome than securing them. Each negative was mounted in front 

 of ground glass illuminated by transmitted light, and was measured with 

 a telescope sliding on a vertical scale. In the focal plane of the eye- 

 piece two spider lines crossed at an angle of 60°, and the intersection 

 of the cross-hairs was brought alternately from above and from below to 

 what appeared to be the middle of the band. At least four settings 

 were made on every band and two settings on the edge of the strip. In 

 this way the height of each voltage and zero band above the edge of its 

 strip was found at six and sometimes seven places of known wave- 

 length varying from 5890 to 4180 Angstrom units, — the wave-lengths 

 being known from the cadmium reference spectrum and the strong 

 carbon bands in the extreme violet. The negative was then inverted, 

 each measurement repeated, and the mean taken to eliminate possible 

 partiality for settings to the upper or lower side of the field of view. 

 How accurately these settings ran may be seen from the following 

 measurements of one plate, — the worst (Tables II and III). How 

 accurately they can be made is shown in Table IV, — perhaps the best. 



In spite of these precautions, reduction and comparison of the meas- 

 urements of eight voltage bands show small discrepancies in the total 

 shift, x, probably due, at least in part, to inaccuracies in the vertical 

 ways on which the telescope was moved. Discrepancies of 0.2 mm., 

 the largest found, might be caused by a very small tilt in the telescope, 



