ON SEISMOLOGICAL INVESTIGATIONS. 



257 



moments. (The period of 21m. itself is, however, obtained in what follows 

 •with remarkable precision to be 21.00155m.) 



The next shock, however, did not come at a whole multiple, but nearer the 

 half-multiple, and seemed to start a new series which continued for eight more 

 shocks, when a third series nearer the original series began to return, and, after 

 a short overlap with the second series, took sole command. 



The three series may be given in tabular form showing simply the differences 

 from an exact sequence of 21m., starting with the original shock as time-zero :— 



Mean 



7-2 



+ 2-2 



Mean +15-7 -I-2-2 



The strongest part of the evidence is the sequence of nine shocks from 

 IV. to XII., after the first series has ended and before the third has begun. 



If we take these by themselves the mean value is +14.8 and the mean of the 

 errors falls to + 2.0m. Vi^hen we have overlapping series there is the objection 

 that we could always choose two points in a cycle of 21m. which would reduce 

 the range of the residuals from + 10.5m., with mean of the errors ±5.2, to 

 ± 5.2m. with mean of the errors ± 2.6m., which is not much greater than those 

 found. But this objection loses its force when we have a consecutive series of 

 nine shocks without departure from the same reference point. 



Let us now consider the following two series of shocks from apparently the 

 same epicentre (6°.0 S. 136°.0 E.), one set in July 1917 and one in Au£u«t 

 1917. ^ 



Date 



Multiple Error 



Date 



Multiple Error 



Mean + 1-5 



Mean +1-8 



In the six cases marked with an asterisk a half-period of 10.5m. has been 

 subtracted from the actual error. The procedure is tolerably clear. In each 

 set the starting point is chosen so as to make the mean of the errors zero and 

 can be recovered from the top case by reversing the error. (The minutes for 

 July 27 are thus 36.2+2.9 = .39.1, and "for August 7, 54.1-1.4 = 52.7.) 



The accident of preparing the bulletins month by month led to the treatment 

 of these two series separately. But we can easily test whether they join together 



T a 



