( 198 ) 



To observ^ation 11 of this table we have applied the correction 

 mentioned at table VI. To the observations 17 and 7 we have 

 accorded half the weight on account of the large deviation from the 

 single determinations mutually (comp. tables VI and VII) '). 



§ 13. Conclusion. For the mean error of the final result for one 

 temperature (when this is taken equal for all temperatures) we find 

 by comparison with the formula found : 



microv. 

 R, ± 2.8 

 R, db 3.2 



R^ dtz 2.6 (2.1 when leaving also out of account — 217°j 

 R^ ±1.8 



The mean error of the result of one day, according to the mutual 

 agreement of the partial results, is : 



±2.9 microvolts, 

 whence we derive for the mean error of one temperature, supposing 

 that on an average two daily results are averaged to one final result : 



±2.0 microvolts. 

 (2 microvolts agree at — 29° with 0°.05, at — 217° with 0°,16). 



Hence it seems that we may represent the electromotive force of 

 the thermo-element constantin-steel between 0° and — 217° by the 

 five-terms formula to within 2 microvolts. For the calibration to — 217° 

 we therefore require measurements at at least 5 temperatures '). 



The representation including the temperatures of liquid hydrogen 

 is much less satisfactory ; for the mean error would be found according 

 to this representation ± 3.2 microvolts, agreeing with 0°.075 at 

 — 29° and 0°.74 at —252° and —259°. 



In order to include the hydrogen temperatures into the formula a 

 6'^ term will therefore probably be required. 



But for measurements at the very lowest temperatures the element 

 constantin-steel is hardly suitable (comp. § 7). 



In conclusion we wish to express hearty thanks to Miss T. C. 

 Jolles and Messrs. C. Braak and J. Clay for their assistance in 

 this investigation. 



1) In the calculations for observations 3, 11 and 5 are used temperatures 0°,081 

 lower than the observed ones. A repetition of the calculation witli the true vakies 

 has not been undertaken, as it would affect only slightly the results, the more 

 because the observations are uncertain. 



2) If the four term formula (comp. footnote 2 §11} should prove for this inter- 

 val as sufficient as the five term formula, this number wouUl be reduced to four. 



