TABLE OF STANDABD WAVE-LENGTHS OF THE SPECTEAL LINES 557 



that this method is far superior to the former. The probable error is, 

 indeed, reduced to Tinnhnnr? or even ^ ess ^ or the best lines. Where 

 the interpolation can be made on photographs, this probable error is 

 scarcely increased at all ; but even taking it at twice the above estimate, 

 the method even then remains from three to five times as accurate as 

 that of angular measurement. Indeed, the impression made on my 

 mind in looking over Vogel's Potsdam observations is, that my tables 

 and process are ten times as accurate as theirs; and I think any careful 

 student of both processes will come to a similar conclusion. 



The wonderful result that can be obtained by the measurement of 

 photographs on the new micrometer, which can measure plates over 

 twenty inches long, is partly seen in the table. Where the distance is 

 only a few inches, the wave-length of a series of lines can be measured 

 with a probable error of less than T ^ of a division of Angstrom. 

 Indeed, a series would determine any line so that the probable error 

 would be even -0000001 of the whole. This would detect a motion 

 in the line of sight of 140 feet per second! 



From the tests I have made on my standards, I am led to believe 

 that down +o wave-length 7000, a correction not exceeding -01 

 division of Angstrom (1 part in 500,000), properly distributed, would 

 reduce every part to perfect relative accuracy. 



To ascend to the next degree of accuracy would need many small cor- 

 rections which would scarcely pay. It is reasonable to assume that a 

 higher degree of accuracy will not be needed for twenty-five years, as 

 the present degree is sufficient to distinguish the lines of the different 

 elements from one another in all cases that I have yet tried. 



DETAILS OF WOEK 



To reduce all the observations in a given region to one line, relative 

 observations extending a short distance either side of the standard 

 region are necessary. Thus the mean of 4215 and 4222 can be taken as 

 the standard, and, if only one is observed, it can be reduced to the 

 standard by a correction -f 3-358 or 3-358. But it is not necessary 

 to take the mean of the lines as a standard, as any one of them may 

 be so taken, or even any other point where there is no line, as the point 

 is only to be used in the mathematical work, and finally disappears 

 altogether. 



Table II gives results of this nature. The letters at the top of each 

 series, e, g, h, ;, etc., are the arbitrary names of the standards. The 

 first columns refer to the series of observations, " Co." being observa- 



