rarely protected from the bending effects of thrusts of trains. What saves these cross-girders from failure is the fact that continuity of the track tends to distribute the thrust over a number of bents. Nevertheless, it is not legitimate to depend on this fact, for, especially on sharp curves, the tendency is to carry the thrust into the ground as directly as possible. In the author's opinion, the only proper way to provide for this thrust is to assume that 20% of the greatest live load between two adjacent expansion points will act as a horizontal thrust upon the columns between these two expansion points; and all parts of the metal-work should be proportioned to resist this thrust properly.

By running a strut from the top of each post diagonally to the longitudinal girder at a panel point of its sway-bracing, the horizontal thrust is carried directly to the post, and a horizontal bending moment on the cross-girder is thus prevented. Such construction should invariably be used where the conditions require it.

XI.   CUTTING OFF COLUMNS BELOW THE BOTTOM OF CROSS-
GIRDERS AND RESTING THE LATTER THEREON.

This style of construction, which until lately was almost universal, is extremely faulty in that there is no rigidity in the connection, and the column is thus made more or less free-ended at the top.

It has been said that no harm is done to the column by making it free-ended, as it can then spring better when the thrust is applied. Unfortunately this reasoning is fallacious, because the few unlucky rivets which connect the bottom of the cross-girder to the top of the column tend to produce a fixed end, and are, in consequence, racked excessively by the thrust of the train. In all cases the column should extend to the top of the cross-girder, and should be riveted to it in the most effective manner practicable.

XII.   PALTRY BRACKETS CONNECTING CROSS-GIRDERS TO COLUMNS.

Brackets are often seen composed of a couple of little angles attached at their ends by two or three rivets. Such brackets are merely an aggravation,  and  are  sure  to  work  loose sooner or later.  Although  it  is

impracticable to compute the stresses in this detail, good judgment will dictate the use of  solid-webbed  brackets  riveted rigidly to both cross girder and col-