THERMO-ELECTRIC DIAGRAM FROM - 200° C. TO 1 00° C. 741 



to — 205°, extrapolated to —283° of platinum temperature (p. 527). This table has 

 been used throughout the following investigation. 



During the intervening years other matters have prevented me from completing 

 this work. It was resumed in the early part of 1908, and is now practically complete. 

 Two remarks have to be made ; first, all the metals examined by Dewar and Fleming 

 are not included here. The alloys I have in general omitted, because their composition 

 is not determined, and we know how an impurity even of a fraction of one per cent, 

 entirely alters many properties of metals. This may be seen on examining the results 

 of Dewar and Kammerlingh Onnes in the resistance of gold within 20° of the 

 absolute zero. For the same reason, I have no results to give for iron. The iron- 

 curve in any case is not a conic. Its appearance, within the range of temperature 

 examined, is somewhat of the form of the old English J ; and it is allowable to say 

 that at present an even approximately pure iron has not yet been produced. The 

 second remark deals with the process of the present investigation. It is entirely 

 experimental. It has therefore been necessary to produce in facsimile the actual curves 

 employed, in order to convey the evidence upon which 1 have felt convinced of the 

 truth of the theory adduced. Had these curves been possessed of such uniformity that 

 one would have sufficed as a specimen, the trouble in proving the theory would have 

 been comparatively light. But this is far from being the case ; in general they are 

 portions of only one side of a (?) parabola, and sometimes the vertex occurs below — 200°, 

 sometimes above 100° ; a few curves contain the vertex within the 300° range of 

 temperature of the observations. Again, when the experimental portion of the curve is 

 very flat, the locus of the middle points of any series of parallel chords which can 

 possibly be drawn is too short to allow it to be stated clearly that the locus is a straight 

 line. Several artifices were tried to meet this difficulty. The temperature, except in 

 two cases which will appear, was measured to the scale of 1 mm. to the degree centigrade. 

 By varying the scale of electromotive force, the curve could be more or less elongated. 

 By too open a scale, the curve would become so long that it would almost appear as a 

 straight line ; but, by going to the other extreme, the magnitude of the curve would be 

 so reduced that even the breadth of the line of the curve would become of importance. 

 A process of "shearing" was finally adopted which may be explained thus : Let PQ, fig. 1, 

 be the experimental A-curve. A straight line BC is chosen somewhat parallel to the 

 chord joining the extremities of PQ, and making with OT, the axis of temperature, an 

 easily determined angle such as tan CBT= 1, or 2, or f. Then the curve and line are 

 "sheared" until BC lies along BT, any ordinate such as RN taking the new position 

 R'N', and PRAQ finally taking up the position P'R'A'Q'. If the new curve P'R'A'Q' 

 should still be too flat, it is only necessary to increase the scale of the new ordinates until 

 a curve of sufficient (but not too great) concavity is obtained. If there is no in- 

 clination made by the axis of PRAQ with OH, the axis of E.M.F., the axis of P'R'A'Q' 

 will likewise show no inclination. The connection between w, the inclination of the 

 axis of the original curve PRAQ to the axis of E.M.F., and u> the angle made by the 



