Serviss — Internal Temperature Gradient of Metals. 455 



were cut and rings of asphaltum varnish spread around them 

 as shown. The weight of the upper cylinder pressed this 

 between the wires, leaving the junctions in narrow annular 

 air spaces in contact Avith the upper cylinder. Possibly it may 

 seem preferable to have placed the rings of additional asphal- 

 tum directly on the thermopile, without the additional mica. 

 But as it was advisable to keep the thermopile flexible, this 

 last asphaltum was dried without baking, hence these thin 

 rings were needed to prevent the cylinders from sticking to 

 the thermopile. If the whole frame had been baked hard, 

 the wires would have been much more liable to accident in 

 changing the specimens, the channels between the wires would 

 not have been so thoroughly closed, and the thermopile 

 would have been more likely to warp, opening a wide gap or 

 producing a strain under the weight of the upper cylinder. 

 Incidentally the inner diameter of the ring in the top of 

 the thermostat had been made slightly too small to admit the 

 thermopile without bending it a little, and by leaving the 

 mica frame flexible it was not necessary to cut this ring. As 

 the distance between the cylinders was not materially increased 

 by the presence of the additional mica, it did no harm, but 

 increased the insulation, distributed the weight of the upper 

 cylinder over a wider area, and also protected the copper wire 

 projecting beyond the cylinders from possible air currents, 

 which might cause trouble. The thickness of the thermopile 

 was - 74 mm and its. resistance at 0° was 48*6 ohms. 



Calibration of the Couples. — 50 cm w T as cut from the middle 

 of each specimen selected for the calibration tests. These were 

 used to make couples as nearly like those on the thermopile as 

 possible, and tested between 0° and 8°. They gave 42 - 2 micro- 

 volts per degree, which agrees very well with the ordinarily 

 accepted value of 43 microvolts. A more careful calibration 

 was not necessary, as the very small differences of temperature 

 which are the subject of this investigation cannot be measured 

 within this degree of percentage accuracy. As it is difficult to 

 maintain small constant differences of temperature for calibra- 

 tion purposes, it is necessary to assume that the thermo-electric 

 height of constantan against copper is a right line in the neigh- 

 borhood of zero. A mean difference of o, 000001 between the 

 inner and the outer row of junctions of the thermopile will, 

 then, generate 4 - 3xl0" 9 volts. 



The thermostat was of thin sheet copper, I7*5° ra in diam- 

 eter and 25 cm in extreme height. A composition casting- 

 was fitted and soldered in the top. The cover was a second 

 casting held to this by twelve machine screws with a gasket 

 between. A light brass ring was suspended by double silk 

 threads, attached, through swivels, to screws (for leveling) in 



