26 Dr. ('. < 'hive. ////-.*///"/ 



r-20 = 2-5, an error of 5 per cent, in the temperature coefficient 

 would introduce the following errors in /: 



Near C. Near 100 C. Near 445 C. 



0-008 0-012 0-OJ.-> 



The above remarks suppose that we are concerned, as in the Kew 

 experiments, with the u/i*ili>ff value of the platinum thermometer 

 resistances, and have to do with observations taken on different days. 

 In ordinary thermometric work uncertainties in the temperature 

 coefficient would be much less serious, supposing one to determine the 

 fundamental interval daily, and to combine together only such observa- 

 tions as are taken on the same day. Under such circumstances it 

 would not matter how far removed from 20" the temperature was, or 

 how much in error the temperature coefficient was, so long as the tem- 

 perature of the coils was strictly constant. Unless, however, the room 

 containing the platinum thermometers is kept at a constant tempera- 

 ture artificially, or is exceptionally situated, its diurnal range of tem- 

 perature will seldom be inappreciable. 



Another source of uncertainty is the possibility of differences between 

 the temperature coefficients of the different box coils, or between those 

 of the two proportional arms. Such differences have not been observed 

 at Kew, but there has been hardly any direct experiment, and the com- 

 parison of the results from two coil combinations, such as CDF and 

 CDGH, in which only the small coils are different, does not afford a 

 delicate test. 



As a difference between the temperature coefficients of the propor- 

 tional arms was suggested by the Cambridge Instrument Company as a 

 possible explanation of some phenomena (due, however, to some other 

 cause), it is presumably a contingency which should not be dis- 

 regarded. 



Error in the temperature coefficients, it should be noticed, would 

 only introduce irregularity into the results of the Kew experiments. 

 It might conceivably introduce an apparent seasonal variation in R) or 

 in the fundamental interval, but could hardly simulate a true secular 

 change. 



Change of Zero or other Error in the Mercury Thermometer inside the Coil 



Chamber. 



24. An error of 0'l C. in the estimated temperature of the coils 

 affects the calculated temperature in the case of one of the ordinary 

 Kew platinum thermometers by 0"'007 at C., 0-009 at 100 C., and 

 0-020 at 445 C. 



Thus for accuracy of the order 0'001 at 100 C., or 0'002 at 445 C.. 

 the temperature of the coils must be known to 0'01 C. An ordinary 



