

Viscosity encountered on passing from Fluid to Solid, 351 



periodically, and thus supply a greater number of instabilities 

 than correspond to the mean twist (position of equilibrium) in 

 question. They may also have a more direct molecular effect 

 in decreasing viscosity. When the rate of twist in the two 

 wires is unequal, viscous motion will also be in excess of the 

 true value. This also applies in case of slightly different 

 lengths, I, of the two effective halves of the wire. In the case of 

 No. 15, which had become somewhat worn by use, the load P 

 probably strained the wire too near the elastic limits. This I 

 infer from the observed successive reduction of viscosities 

 from twist to twist, clearly indicating that an excess of insta- 

 bility is being supplied. It is therefore better to accept the 

 behaviour of No. 2 (soft) as typical of No. 15 ; for the normal 

 viscosities of the two states of temper are identical. 



Discussion. 



15. Turning first to the values of <£, it is seen at once 

 that they substantiate the inferences of my earlier papers 

 throughout*. They therefore need no further comment here. 

 Tables V., VII. contain data for my normal wire, by aid of 

 which the differential results of the earlier papers may be 

 reduced to absolute values. I have endeavoured to make the 

 angles t of nearly the same value in all cases, so that <J)/t 

 may be comparable. 



16. The spontaneous breaking of the glass-hard wire in 

 Table VI. deserves mention. It occurred during the third 

 twist, and at a time when the wire was in no way interfered 

 with. This corresponds to the spontaneous explosion of hard 

 projectiles frequently observed. It also corresponds to the 

 spontaneous rupture of stressed glass, whether the stress be 

 stored internally or applied externally. I have observed this 

 interesting phenomenon in glass under a great variety of 



* Phil. Mag. xxvi. p. 183 (1888) ; ibid, xxvii. p. 155 (1889). 



In the deductions of the former paper (p. 189) I neglected the elastic 

 motion, in virtue of which, at every stage of viscous yielding, the rate of 

 twist is maintained constant throughout the system. After correcting 

 for this, 



L^=2/'{^ 3 0-0 + ^(o-/3) + ^) 1 (L-a)}-2^T; 



which under the simplified conditions of experiment becomes \//- = /(<£ — </>'). 

 Hence it follows that the data {(f) — cj)') of the two papers cited are 

 relative, and must be multiplied by ln2 to reduce them to the dimensions 

 defined in the text. Beyond this the error is without importance, the 

 data applying at once to rods initially somewhat harder than those 

 actually observed, 



