134 Further experiments on the (Marcu, 
The error of the readings therefore upon a length of 10 feet (assum- 
ing it even to a 23 divisions of the micrometer) will not surpass 
0.000001, while the error of the thermometer reading may amount 
to 0.000002: it will be seen from the tables which follow, that the 
general run of the experiments fully confirms this estimate of accuracy ; 
at the same time it would be useless to carry the expression of the 
dilatation beyond the sixth decimal, as is frequently done in cases less 
entitled to reliance. ; 
The order of each experiment was similar to that described in my 
former paper. When several readings had been made at the temper- 
ature of the room, the steam was let on and kept up for several hours, 
during which the second readings were made. Cold water was not 
introduced, as it took a long time to restore an equal temperature, and 
it was found better to allow the apparatus to cool down gradually by 
the following morning. 
It was only in the third series of experiments that the bar remained 
quite stationary at the higher temperature for more than two hours. In 
general it was remarked that the reading of the micrometers gave the me- 
tala maximum dilatation at the first moment of its being brought to the 
boiling point, gradually falling off even to the extent of 20 divisions 
(zccc0th of an inch), as the steaming continued. This was evidently not 
attributable to change of temperature in the steam, for the thermometers 
were not affected. I imagined that it must be produced by torsion or 
curvature of the bar, from the under part of it being at first less heated 
than the upper ; for, by the construction of the apparatus for steaming, it 
is evident that, on the introduction of the steam, the upper parts of the 
tube would become heated first, while the condensed steam collected in 
the lower part of the cylinder imparted a lower temperature to the 
under surface of the bar: but this would cause the bar to assume a 
slight curvature upwards, which, as the supporting rollers were situ- 
ated in distance one-fourth from the ends, would tend to depress the 
dots below the true focus of the microscope; the effect of this and of the 
curvature would be to make the bar shorter than otherwise, so that this 
explanation cannot be admitted. 
Some very curious experiments, however, which are described by 
Captain Kater in the Phil. Trans. for 1830, may serve to afford an 
explanation of the anomaly. That gentleman found that the error in the 
linear measurement of a flat bar of 36 inches in length, might amount 
to .001 inch, simply by its resting upon an uneven surface, and assum- 
ing a curvature therefrom, the versed sine of which amounted to no 
more than .01 inch; now the difference between the chord and the’ 
