DISCUSSION OF BAROMETRIC OCSERVATIOXS. 



467 



a contrary opinion may be derived by examining the deviation of each 

 l^eriod from the normal pressure. 



As will presently be shown, the i^robable error of one of the values in 

 Table III is a variable quantity, having for its mean value 0.0095, while 

 at its maximum it is about 0.0122 and at its minimum 0.0058. If now 

 we compute a value for the middle day of each period from the periodic 

 function, in the same way as Table I, but to four places (which may be 

 regarded as representing accurately enough for the present purpose the 

 normal value of pressure for the period, aside from the irregularities now 

 under discussion), these normal values, subtracted from the successive 

 values of Tables III, will leave the following residuals : 



1 + .0158 I 19 + 



2 —.0260 20 — 



3 —.0096 21 + 



4 +.0131 22 — 



5 +.0012 23 + 



6 —.0081 24 — 



7 4- .0018 25 + 



8 + .0081 26 — 



9 — .0510 I 27 + 



10 + .0317 ' 28 — 



11 + .0U41 I 29 — 



12 



13 



14 + .0178 I 32 



15 + .0043 : 33 — 



16 — .0038 I 34 — 



17 — .0055 35 + 



18 — .0015 36 — 



+ .0040 i 30 + 



— .0229 I 31 + 



+ 



. 00(;5 



. 0183 

 . 0020 

 . 0180 

 . 0072 

 . 0104 

 .0255 

 . 0015 

 .0048 

 . 0318 

 . 0291 

 . 0421 

 . 0253 

 . 0016 

 . 0148 

 . 01.59 

 . 0310 

 . 0002 



It here appears that the periods following the March and May minima 

 and the January maximum deviate from the normal value in the oppo- 

 site direction from the adjacent critical value, and to an amount 

 exceeding the estimated probable error for the time of year, while the 

 periods preceding the January and September maxima appear to be 

 normal, the double maximum in these cases being j)roduced by a rela- 

 tively high value of the period in which the anticipatory subordinate 

 maximum occurs, rather than by a relatively low value in the period 

 separating it from the true maximum. The effect of the disturbance in 

 March- April appears to be confined to the ninth and tenth periods, and 

 it produces that divergence of the computed and observed means for the 

 two months, previously noticed in connection with Table I. 



The division of the year into thirty-six equal i^eriods, in Table III, 

 gives the opportunity for a recomputation of Bessell's periodic function. 

 The following are the coefficients of ten terms, based upon the thirty-six 

 periodic means. The first five or six terms may be regarded as replac- 

 ing by more accurate values the formula previously derived, and from 

 which Table I was computed ; the concluding terms are, perhaps, chietiy 



