( 87 ) 
Throughout the preceding investigation it was assumed that the 
influence of the temperature is proportional to its first power. It is 
important to investigate the results which will be found, if we repre- 
sent the influence of the temperature by the formula: 
e, (t{—t,) + ¢, (¢—t,)’. 
For this purpose I used the deviations of the monthly means 
from the values of a derived from the curve. These deviations were 
represented by the formula 
Aa + Ae, (t—t,) He, (tt). 
I did not investigate the separate vears, but I derived mean results 
for the three periods mentioned above; f, is then in each case the 
mean temperature of the period, and differs but little from —+ 8.7 
(= + 8.1 Réaumur). 
In this way I found the following values of Ac, and c,, both 
expressed in tenthousandth parts of a second: 
Ae C5 
19792=1964° 1-30 439 
Ke EE 
1893—1897 19 —7.9 
The values of Ac, nearly agree with those previously found 
for Ac. Those of c, are small and of different sign, and their reality 
is doubtful. The rates for temperatures below zero would require 
positive and much larger values of c,. In order to represent e. g. the 
two results for the months 1890 Dee. and 1891 Jan. it would be 
necessary to assume c, = +15. 
I think therefore to have acted correctly by exeluding the rates 
corresponding to temperatures below 0°. For the other temperatures 
we may certainly provisionally adopt a linear formula for the influence 
of the temperature. 
As to the coefficient c of this formula, I do not think that it could be 
represented as a function of the time which would have any real 
meaning. Probably, however, it will be better to assume if constant 
during shorter periods only, e. g. thus: 
Ac C 
1819-801121; AAT 
1881—83 -+ 52 — 216 
1884 +31 — 237 
1885—91 ste 
189093.) Sir oad 
(Bode IT EB, aay 
Finally [I will show how these coefficients would be altered (see 
