ay, by causing the train of wheels to lift up a certain
specified weight, and let it drop on the pendulum at regular intervals,
or by some equivalent method.

The two requirements above stated have given rise respectively to what
are known as detached escapements, and remontoires, which will be
described presently. In the first place, however, it is desirable to
describe the principal forms of escapement in ordinary use.

Balance escapement.

The balance escapement, which has been already mentioned, was in use
before the days of pendulums. It was to a balance escapement that
Huygens applied the pendulum, by removing the weight from one arm and
increasing the length of the other arm.

[Illustration: FIG. 8.--Anchor or Recoil Escapement.]

Anchor escapement.

Very shortly afterwards R. Hooke invented the anchor or recoil
escapement. This is represented in fig. 8, where a tooth of the
escape-wheel is just escaping from the right pallet, and another tooth
at the same time falls upon the left-hand pallet at some distance from
its point. As the pendulum moves on in the same direction, the tooth
slides farther up the pallet, thus producing a recoil, as in the
crown-wheel escapement. The acting faces of the pallets should be
convex. For when they are flat, and of course still more when they are
concave, the points of the teeth always wear a hole in the pallets at
the extremity of their usual swing, and the motion is obviously easier
and therefore better when the pallets are made convex; in fact, they
then approach more nearly to the "dead" escapement, which will be
described presently. The effect of some escapements is not only to
counteract the circular error, or the natural increase of the time of
a pendulum as the arc increases, but to over-balance it by an error of
the contrary kind. The recoil escapement does so; for it is almost
invariably found that whatever may be the shape of these pallets, the
clock loses as the arc of the pendulum falls off, and vice versa. It
is unfortunately impossible so to arrange the pallets that the
circular error may be thus exactly neutralized, because the escapement
error depends, in a manner reducible to no law, upon variations in
friction of the pallets themselves and of the clock train, which
produce different effects; and the result is that it is impossible to
obtain very accurate time-keeping from any clock of t