4'J I>r. ( '. Clnvr. / on 



more weight to the result of the first calculation. In any case I have 

 little doubt that the formula 



* = /, + 0-090 (^-760) + 0-0002 (^-760) 2 ......... (17) 



will prove considerably more exact than the formula (10) now in use. 

 Further direct experiments on the point are, in my opinion, desirable. 



39. In our previous calculations we have employed Callendar and 

 Griffiths' value, 444 -53 C., for the boiling point of sulphur. This is 

 supposed to be measured on the scale of the constant pressure air 

 thermometer. It is about 0'7 C. lower than the value recently found 

 by Chappuis and Harker, employing the constant volume nitrogen 

 thermometer. We do not as yet know enough about gas thermometry 

 to judge of the consistency of the two results, or of the degree of 

 accuracy either can claim. It is thus desirable to know how the 

 previous results would be influenced by a slight change in the accepted 

 value of /. Such a change would not affect any formula like (10) con- 

 necting / and p, and it would modify formulae like (13) containing^ 

 only indirectly through change in 8. The change AS in 8, answering to 

 a small change A/ g in t f , is given by 



AS = A/. [10 4 - 8(2*, - 100)] -=- [/,(*, - 100)]. 



Using Callendar and Griffiths' value of t g we get approximately, 

 when 8 = 1-5, 



AS = 0-0575A/,. 



For instance, if A/ x = 0'7 C., 



then AS = 0'04, approximately. 



Constancy oi" Variability of Plot i/nim Thermometer*. 



40. Supposing the RQ of a thermometer, determined in the usual 

 way, to show a distinctly lowered value as time proceeds, the cause 

 might be a change in the leads, or in the proportional arms, or in the 

 box coils, and not a real change in the thermometer itself. There are 

 no certain data as to the original equality of the proportional arms ; 

 but as Dr. Harker recently found them to differ by only about three 

 parts in 20,000, and as they were, doubtless, made as nearly as possible 

 equal at first, there is reason to suppose that any uncertainty on this 

 head is small. 



On the other hand, as shown by Table III, changes in the leads have 

 been large and fluctuating. As no data exist for eliminating this 

 uncertainty in a direct way in the earlier observations, I have based 

 the inquiry into the constancy of the thermometers mainly on an inter- 

 comparison of themselves. As the same coil combinations were used 

 for every one except K<j, and as the thermometers have nearly equal 





