can
be readily and accurately tested. The absolute value of the E.M.F. of
the Clark cells employed was determined with a special form of
electrodynamometer (Callendar, _Phil. Trans._ A. 313, p. 81), and
found to be 1.4334 volts, assuming the ohm to be correct. Assuming
this value, the result found by this method for the specific heat of
water at 20 deg. C. agrees with that of Rowland within the probable
limits of error.

S 15. _Variation of Specific Heat of Water._--The question of the
variation of the specific heat of water has a peculiar interest and
importance in connexion with the choice of a thermal unit. Many of the
uncertainties in the reduction of older experiments, such as those of
Regnault, arise from uncertainty in regard to the unit in terms of
which they are expressed, which again depends on the scale of the
particular thermometer employed in the investigation. The first
experiments of any value were those of Regnault in 1847 on the
specific heat of water between 110 deg. C. and 192 deg. C. They were
conducted on a very large scale by the method of mixture, but showed
discrepancies of the order of 0.5%, and the calculated results in many
cases do not agree with the data. This may be due merely to deficient
explanation of details of tabulation. We may probably take the
tabulated values as showing correctly the rate of variation between
110 deg. and 190 deg. C., but the values in terms of any particular
thermal unit must remain uncertain to at least 0.5% owing to the
uncertainties of the thermometry. Regnault himself adopted the
formula,

s = 1 + 0.00004t + 0.0000009t^2 (Regnault), (3)

for the specific heat _s_ at any temperature _t_ C. in terms of the
specific heat at 0 deg. C. taken as the standard. This formula has
since been very generally applied over the whole range 0 deg. to 200
deg. C., but the experiments could not in reality give any information
with regard to the specific heat at temperatures below 100 deg. C. The
linear formula proposed by J. Bosscha from an independent reduction of
Regnault's experiments is probably within the limits of accuracy
between 100 deg. and 200 deg. C., so far as the mean rate of variation
is concerned, but the absolute values require reduction. It may be
written--

s = S_100 + .00023(t - 100) (Bosscha-Regnault) (4).

The work of L. Pfaundler and H. Platter, of G.A. Hirn, o