So again, it is certain that a considerable part of the peripheral drainage of glaciers takes place through tunnels beneath the ice. It is reasonable to suppose that during the winter season, when the drainage is slack, these tunnels tend to collapse in greater or less degree, under the continued pressure of the ice and the "fattening" of the glacier, so that in the early part of the next melting season the contracted tunnels may be over-flooded by glacial waters. To the extent that these tunnels become incompetent the water would become ponded back in the crevasses and moulins by which the surface-water gains access to them. They thus come to have something of the force of water flowing in tubes, and may be presumed to be capable of forcing rounded material to some considerable height, and of carrying ice-imbedded boulders to any point reached by the stream. These tunnels probably undulate with the bottom, and lodgment along them takes place wherever enlargement permits. Without, therefore, appealing to any upward cross currents within the ice itself, it is possible to explain the transportation of the drift from lower to higher altitudes. I have never seen phenomena of this kind that seemed to call for any other explanation than these. I am not prepared to say that there are no such phenomena. One of the purposes of this article will have been accomplished, if it shall call forth a critical statement of phenomena that require the assumption of internal upward movements of the ice to account for them, and of the criteria which distinguish such phenomena from those that may be referred to upward basal movements such as are common to all streams or to the exceptionally conditioned subglacial streams. That there are upward internal movements in most streams is as much beyond question as the existence of upward basal currents in rivers and glaciers, but they are dependent chiefly upon the velocity of the current and the irregularity of the bottom. Theoretically, as I understand, a stream moving in a straight course on a perfectly smooth bottom would not develop an upward cross current. Each lower layer would move slower than that above it by reason of basal friction, but they would move on in parallel lines. But if irregularity of bottom be introduced the parallelism is obviously destroyed, and if the velocity be high so that the momentum of the particles becomes great relative to their cohesion, irregular internal movements will result, and these will often be of a rotary nature in vertical planes bringing the basal parts of the fluid to the surface or the reverse. For this reason rapid streams abound in rotary currents, while slow streams do not.