Pellet injection is a primary method of replenishing density loss in large size, long pulse tokamaks, such as ITER, which are intended to operate for many particle confinement times . The method has been tried in many experiments, and is generally considered successful, although a significant fraction of the pellet mass can be lost, not absorbed into the plasma.
The location at which pellets are injected has an effect on how much of the pellet mass is retained in the plasma. Usually, pellets are injected from the outboard, large major radius edge of the plasma, because this is the most accessible part of the experimental apparatus. Recent ASDEX experiments  demonstrated that pellets injected on the inboard, low major radius edge, suffered less loss, and were absorbed more completely into the plasma.
For very large tokamaks, such as ITER, there is a further requirement to get pellets to ``fall into" the center using this mechanism, because they will ionize relatively near the plasma edge.
The cause of this effect is the toroidal curvature, the same effect responsible for the outward Shafranov shift of the flux surfaces, acting in conjunction with expansion of the ionized pellet ablation cloud. We demonstrate the effect using MHD simulations.
The numerical results confirm that pellets injected at the outboard edge lead to an MHD equilibrium in which the injected density moves relatively closer to the outboard plasma edge. We also show that sufficiently massive pellets, injected at the inboard side of the plasma, are able to reach the plasma center. In the process, the pellet drives magnetic reconnection leading to a reversed shear profile. This may offer a way of maintaining reversed negative central shear, which can be highly favorable for transport, in the presence of Ohmic dissipation in large long pulse or steady state tokamaks.
Finally, pellets injected at the top (or bottom) edge lead to an MHD equilibrium in which the injected density moves relatively little.
We derive a simple model of the pellet displacement, which agrees well with the numerical data obtained from the 3D simulations. The pellet displacement is proportional to the pellet pressure perturbation, and the cosine of the angle between the normal to the flux surface at the pellet and the major radius direction.