520 PROFESSOR DITTMAR AND MR C. A. FAWSITT ON 



in an earlier part of this paper, we always worked with two samples of alcohol 

 at the same time, and these determinations included a complete rehearsal of 

 the determination of the entire tension-curve. But in the final series we 

 preferred to charge only one limb to reduce the number of readings by one. 

 The actual routine of the work hardly requires to be described. After 

 having established the desired temperature, we read first the three tensiometer 

 limbs, then those of the manometer, and, lastly, the thermometer a second 

 time, by means of a horizontal telescope. The temperature was in all cases 

 rigorously constant, as far as one could read. To eliminate part of the error 

 arising from unavoidable variations of temperature during a series of read- 

 ings, we found it an improvement to close the open end of the manometer, 

 and thus fix its mercury-menisci in their positions, immediately after reading 

 the limbs in the tensiometer. As a rule, we commenced with the lowest 

 temperature to proceed step by step to the highest, and then retraced our 

 steps, so that each series consisted of an ascending and a descending section. 



The alcohol used for the final tension determinations was specially dehy- 

 drated (by means of CuS0 4 ), and a sufficiency kept in a sealed-up tube until the 

 tensiometer was ready for its reception. 



The height of the several mercurial columns was reduced to 0° C, but not 

 reduced to any standard latitude, for an obvious reason. 



A preliminary survey of the results showed that they fell in approximately 

 with the equation 



log p = a + bt . 



We accordingly for a first approximation adopted this function, determined the 

 constants a and b graphically, and from them calculated the values log p, 

 corresponding to the several observational fs. For a second approximation we 

 laid down the ^s as abscissas, and the corresponding values, 



"Al/"~{^ogp as observed) — (log p as calculated), 



as ordinates in a system of rectangular coordinates, when Ay appeared to be a 

 function of t, according to an equation of the form Ay = a + fit+yt 2 . We then 

 calculated the constants a, /3, and y from three measured ordinates, and thus 

 established an equation, 



log p — a + b't + ct 2 + (5(log p) , 



where S(log p) stands for the residual correction needed to establish equality 

 between the two sides of the equation. These residuals 8(log p), when repre- 

 sented graphically in function of t, suggested an equation of the 4th or 5th 

 degree ; but on looking more critically into the matter, we found that this final 



