its Relations to Temperature. 3 



To heat the upper wire, a given part of it was surrounded 

 by a special form of vapor bath, without being necessarily in 

 contact with it. This will be described elsewhere. Steam 

 (100°), aniline vapor (190°), and mercury vapor (360°) were 

 consecutively introduced. Errors due to radiation are care- 

 fully avoided by appropriate screens. 



By removing and then re-inserting the pin q, the wires may 

 be twisted in multiples of 90° for each length, or in multiples 

 of rate of twist (r) of 3° for each wire. For the given dimen- 

 sions this is the effect of a couple of 250 g. on one centimeter, 

 for each 3° of rate of twist. 



A few critical remarks on the efficiency of the apparatus are 

 in place. In most of the experiments made the torsional de- 

 formation is so great that special fiducial marks to register the 

 possible motion of the fixed pieces gh and hi are superfluous. 

 Indeed, for angles as large as those observed, Grauss' method of 

 angular measurement is no longer conveniently applicable, be- 

 cause of the number of corrections which become essential. 

 In future measurements it will be expedient to use other meth- 

 ods. The large deflections, however, made it possible to use 

 the Gauss' method even when the apparatus is thrown into 

 unavoidable vibration ; for instance, when the upper wire is 

 surrounded by boiling mercury. In the present work it did 

 not seem necessary to take special precautions for jacketing the 

 lower wire. Its temperature is that of the surrounding air, 

 very nearly, and it is not heated by radiation. In the case of 

 steam and of aniline vapor, the length of the hot part of the 

 upper wire is sharply measurable, but in the present work this 

 could not be so satisfactorily done for mercury vapor. Begard- 

 ing the mode of heating, there are two methods available : the 

 wire may either be heated and the twist then applied, or the 

 wire may be twisted to the desired amount before heating. 

 The former of these methods eliminates the time error, but it is 

 difficult to obtain an accurate reading for the zero-point, i. e., 

 the scale reading for the twisted system before heat is applied. 

 Hence the other method was adopted, in which the invariability 

 of the zero referred to can be satisfactorily tested before each 

 experiment. The error is the time consumed in heating the 

 wire, and in the second part of the observation, the time con- 

 sumed in cooling the wire after heating. Special means for 

 cooling were not applied. "When the motion is great, it is de- 

 sirable to have the index-mirror adjustment light, though the 

 device used was not perhaps objectionably heavy. 



The immediate effect of heating the upper wire is an expan- 

 sion of the system. Hence if the part of the low fixed brass 

 M, along which the pin q is free to slide, be not plane and true, 

 there will be a tendency to rotate the system. If, however, the 



