1894.] Specific Heat of Water in Electric Units. 27 



as given by Kahle (' Zeits. f. Instrumentenkunde,' vol. 13, p. 310, 

 1893). Glazebrook and Skinner's coefficient refers to a mean tempera- 

 ture of 7*5, and is identical with the above at that temperature. 



The comparison of the results of different observers will be facili- 

 tated by Table I, in which we compare the work done in ergs with 

 the foot-pound at Greenwich and the kilogrammetre at Paris. This 

 table has been calculated on the assumption that g at Paris is equal 

 to 980*96, and at Greenwich equal to 981*24. We have prepared 

 another table (II), which at any temperature will give the correc- 

 tion of an interval measured on our mercury thermometer to an in- 

 terval measured on the nitrogen and hydrogen thermometers. This 

 table has been calculated with the help of the equation given by 

 Chappuis for the correction to the thermometer made of French hard 

 glass. 



In comparing our results with that of other observers, we have in 

 the first place to consider the value which Mr. Griffiths has obtained 

 in his very excellent series of measurements. His final result 

 (' Hoy. Soc. Proc.,' vol. 55, p. 26 ; ' Phil. Trans.,' clxxxiv, A, 1893) is 



J = 4*1982(1-0*00266 015) x 10 7 . 



This refers to the nitrogen thermometer. At a temperature of 19*1, 

 the value would be reduced to 4*1936, which corresponds to our 

 4*1905 at the same temperature. Griffiths' value is to be increased 

 slightly, owing to the fact that he really measures the difference 

 between the specific heat of water and of air. This would increase 

 the value of J by *0011 about, so that the value of J at 19*1 would 

 be raised to 4*1947 x 10 7 , which is exactly one part in a thousand 

 larger than ours. The difference is small, but must be due to some 

 systematic error, as both Griffiths' value and our own agree so well 

 with each other, that ordinary observational errors and accidental 

 disturbances could not have produced so large a difference in our 

 results. The least satisfactory part of a calorimetric measurement 

 must always be the application of the cooling correction, and we have 

 considered it of great importance to reduce that correction as much 

 as possible. The uncertainty of the cooling correction does not 

 necessarily depend on its value ; thus we can much diminish it by 

 starting, as -we have done in the third series, with the initial tem- 

 perature of the calorimeter about as much below that of the water 

 jacket as the final temperature is above it ; yet the uncertainty of the 

 correction does not seem to us to be diminished by that process. We 

 may reasonably estimate the uncertainty due to the cooling correc- 

 tion, by calculating what the error in the observed rate of cooling, 

 either at the beginning or the end of the experiment, nmst have been 

 in order to produce a difference of one part in a thousand in the 

 final result. We find in our own experiments that the error must 



