These figures are probably quite safe enough even for the most important works and for those where the heading up is constant. For small works or for regulators (_Art. 5_) where the heading up is not constant, steeper gradients are permissible. Much also depends on the condition of the water. If it contains much silt, all interstices will probably become choked up. The hydraulic gradient in the case of the Narora weir across the Ganges was 1 in 11. The weir failed after working for twenty years. It was rebuilt with a gradient of 1 in 16. In the Zifta and Assiut regulators on the Nile the gradients are 1 in 16·4 and 1 in 21. [Illustration: NARORA WEIR AS ORIGINALLY BUILT.] [Illustration: NARORA WEIR AS RECONSTRUCTED.] [Illustration: FOUNDATION OF THE ZIFTA REGULATOR, RIVER NILE.] Regarding the upward pressure on the floor due to the hydrostatic pressure from the head A B, there is a theory that the weight of a portion of the floor at any point P should be able to balance the pressure due to a head of water P R. This, supposing the masonry to be twice as heavy as water, would give a thickness of floor equal to half P R. According to Bligh, the theoretical thickness ought, for safety, to be increased by one-third. Practically the thickness need not, in most cases, be made even so great as is given by the theoretical rule. On canals in the Punjab it is certainly less. Water passing through soil or fine sand does not exert anything like the pressure which it exerts when passing through a pipe. It acts in the same manner as in a capillary tube. It is only in coarse sand or gravel or boulders that water flows as in a pipe.[12] If the tail water covers the floor, the weight of a portion of floor is reduced by the weight of an equal volume of water. If the foundation of any part of the floor is higher than B E, the upward pressure on it is reduced because the water has to force its way upwards through the soil. Bligh also states as an empirical rule that in order to provide efficiently against scour the length of floor B E should be 4/_s_ √(H/13), where H is the maximum head A B; and he points out that in a case where this length is less--as it usually is--than that necessary to give a hydraulic gradient of the requisite flatness, according to the rule previously quoted, it is better to add an upstream floor B D, which may be of puddle and therefore cheap, than to add to the downstream floor a length E C which must be of masonry or concrete, and that this arrangement, by shifting the line of hydraulic gradient from A E to F E, gives a reduced upward pressure on the downstream floor. The length E N to which pitching, if of “rip-rap” type, should extend is given by Bligh as 10/_s_ √(H/10) √(_q_/75), where _q_ is the maximum discharge in cubic feet per second passing over a 1-foot length of the weir, and H is the head A B.