New Standards of Optical Wave-length. 369 
visible spectrum, and showed that Rowland’s system con- 
tained values alternately too high and too lew. It is now 
quite certain that Rowland overestimated the correctness of 
his measurements, that by using his numbers as standards 
we never can be sure of obtaining relatively exact numbers 
within 0:01 A. Besides these fluctuating errors pointed out 
by Perot and Fabry, many of Rowland’s lines have individual 
errors amounting to | or 2 hundredths, as any one may see 
by using different standards for the determination of the 
same line. Some years ago I tried to give a table of iron 
standards free from these individual errors, but of course my 
table as based on Rowland’s is subject to nearly the same 
fluctuations of values. 
For the practically more important ultra-violet part of 
Rowland’s system we have no evidence whatever of correct- 
ness or incorrectness; but it is highly probable that it will 
contain errors of the same kind as the visible part. 
If one tries to determine the wave-length of a sharp line 
—and nearly 99 per cent. of all spectrum lines can be 
obtained sharp—by interpolation between neighbouring 
standards, he will not find it difficult, by using a good con- 
cave grating and a good dividing-engine, to get with a single 
measurement the value within five thousandths of an A, if he 
uses the same standards; but by using others he may get 
differences of one or two hundredths. So it appears that the 
uncertainty of our wave-length determinations is caused by 
the insufficiency of Rowland’s standards. 
To insist upon the importance of obtaining wave-lengths 
as correct as possible for astronomical purposes, for the iden- 
tification of elements, for the series relations, &c. seems 
needless. So the necessity of creating a new system of 
standards is evident; and it seems time to consider the way 
by which a better system may be produced and the errors 
committed by Rowland avoided. 
One of Rowland’s errors was the use of mixed solar and 
arc lines which he thought were of identical wave-length. 
Since that time we know, by the publications of Jewell, 
Humphreys, and Mohler, that they are different. So this 
cause of errors could be easily avoided, and it seemed as if 
by applying the method of coincidences only to arc-lines a 
correct system could be produced. During the last twelve 
years I have three times tried to do so, but have always 
failed,—I did not know why. 
After the publication of MM. Perot and Fabry’s measure- 
ments of iron lines, I formed the plan of founding the method 
of coinc dences on their measurements; it is much easier to 
Phil. Mag. 8. 6. Vol. 8. No. 46. Oct. 1904. 2R 
