570 Prof. H. Kayser on New 
get a system of standards throughout the spectrum by using 
many fundamental lines than by using a single one. But as 
I was not sure of the absolute correctness of Perot and 
Fabry’s numbers, though they seemed very reliable, I felt 
obliged to convince myself. This can be done by the method 
of coincidences, as one line can be got by that method from 
others. 
While engaged in this research, there appeared a paper 
from Michelson showing that errors in the ruling of gratings 
are possible, which make the method inapplicable : : a line of 
given wave-length in the first order does not coincide with 
another line of half that wave-length in the second order. 
I immediately changed the scope of my investigation, and 
made some tests to see whether Rowland’s gratings have such 
anerror. I have two of Rowland’s lar gest oratings ruled on 
his second and third div iding-engines. The comparison of 
results obtained with them by the method of coincidences 
gave differences of more than 3 hundredths A, while repeated 
determinations with each grating alone agreed to a few 
thousandths. 
It is thus apparent that Rowland’s gratings are not suited 
for the application of this method, and it is probable that the 
larger part of the errors in Rowland’s system are due to this 
cause. As gratings are then not to be used for exact 
absolute determinations, they are useless also for relative 
measurements, if great accuracy is desired. At present we 
have no better engines, though different American physicists 
are planning new constructions, as Michelson and Wadsworth: 
time must show if their engines are better. We are now 
under the deplorable neces sity of disregarding the method of 
coincidences in the creating of a new ~ system of standards; 
the spectra of gratings can be used only for interpolation 
between standards not too far apart. 
It seems to me, then, that the only way open consists in 
the determination of a great number of arc-lines with some 
interferometer ; and probably Fabry and Perot’s form of 
that instrument will prove most appropriate. Its use is not 
easy, it requires long practice even for the visible part of the 
spectrum, and for the ultra-violet part the difficulties will be 
much greater. Not a single observer, but several, should 
undertake this task, and when their numbers agree to one or 
two thousandths A for, say, fifty or a hundred lines throughout 
the spectrum, then again the grating may be applied for 
getting by interpolation some hundreds of lines more. Then it 
will be possible to get all the spectral lines accurate to some 
