158 



EEPORT — 1882. 



(14) Perhaps, however, more is to be learnt from the curves actually- 

 obtained than from purely theoretical discussion. 



Unless the greatest care is exercised, the errors due to ' taste ' in 

 drawing the curves may far exceed those produced by mere instrumental 

 inaccuracies. 



An excellent example of this is afforded in the neighbourhood of the 

 point 135° on Thermometer C. 



Figure 2, Plate I., gives a part of the curve obtained by the Gay- 

 Lussac method with the thread l°-6 long. The points surrounded by 

 circles are those obtained by the thread 4° long. The dotted line is that 

 actually drawn b}'' a person unacquainted with the form of the other 

 curve. Thus at 138° a difference of 0°-012 is introduced. A similar 

 result was obtained with the two Gay-Lussac curves shown in fig. 1. 

 The dotted curve happened to have a point at 135° where the correction 

 ■was found to be very low, in this agreeing with the Bessel curve, fig. 3. 



tsessEL 

 5 



10 



The other curve missed this point, and gave a much higher correction 

 at 136°. Probably both are nearly right. There is evidently some sudden 

 change or irregularity at this part of the scale, to fully investigate which 

 a number of points taken very close together would be required. 



If a thermometer is required to be accurate to 005 m.m. (0°'005) 

 points must be determined at intervals not greater than 20 m.m., and 

 wherever any sudden bend in the curve occurs, a secondaiy calibration 

 of that part of the scale should be instituted with a shorter thread. 



(15) The following Table (V.) gives the corrections for the principal 

 points as determined by all the methods. Rudberg's is omitted, as these 

 points could not be directly corrected by that method. The corrections 

 obtained by it are, however, plotted down in fig. 3, Plate 1. 



(16) A very careful comparison has been made between the results 

 given by Bessel's and the other methods. The two methods of calculation 

 applied to it are fully discussed in Part II. The result is that whereas 

 at the end of the second approximation the curves disagree by nearly 



I 



