132 REVERSED AND NON-REVERSED SPECTRA. 



varies in different, otherwise satisfactory experiments; or the failure to com- 

 pletely exhaust the tube may leave a small error which becomes appreciable 

 in B. 



77. Conclusion. If allowance is made for the fact that AN at the 

 micrometer is measured for air, at barometric pressure p' and absolute tem- 

 perature T, the equation for ju i at normal conditions would be 



_ _AAT 76 T i 2B 



~e~~$^n i-p'&N/ep "X*" 



where the correction factor iANp'/pe would not appreciably modify the 

 results. 



It is difficult to see, therefore, why the promising results of 4, which are 

 quite as near the standard data as the method warrants, did not bear consis- 

 tent fruit in the sequel. The direct values of B X io 14 are usually too small, 

 sometimes too large, and range from i to 2. On the other hand, MO i usually 

 comes out too large, whereas it should be correct to a few tenths percentage. 

 None of the causes examined, temperature, pressure, thermal coefficient, 

 flexure of glass, etc., quite account for such a result. If BoX io 14 is computed 

 from standard results for HQI and observed at different spectrum lines, the 

 data are nearly correct for some lines, but too large for other lines, so that a 

 single constant does not reduce the series. It does not seem probable, however, 

 that equation (i) is inadequate; for the results obtained with equal care at dif- 

 ferent times for the same H Q i or B are not in accord. The discrepancies, in 

 other words, are not persistent in value and are therefore due to some inci- 

 dental cause which has not been detected. It has seemed to me that the change 

 in shape of the interference pattern on passing from red to violet, which in 

 case of ordinary glass mirrors is marked, may be responsible for some of the 

 difficulties encountered. This pattern, which for optically flat surfaces would 

 remain elliptical, becomes more and more irregular as the distances, e, of the 

 mirror and grating are increased. The distorted image shrinks laterally from 

 red to violet fully one-half, so that it is not certain that the center of figure 

 is actually a fiducial point. The question, however, would have to be tested. 



