What makes this magnesium blockade of the NMDAR channel particularly significant in terms of LTP induction is that the block is membrane voltage-dependent. The basis of this voltage dependence is relatively straightforward. The NMDAR channel is a transmembrane protein; that is, it spans the cell membrane. As such, it also spans the electric field generated by the membrane potential. The magnesium binding site within the NMDAR channel is physically located within this electric field. Magnesium ions carrying a double positive charge can be acted upon by the field. When the cell is hyperpolarized, magnesium is stabilized inside the channel (i.e. the two positive charges on the magnesium ion are attracted toward the negative pole of the electric field, which points toward the inside of the cell). As a cell is depolarized, the field effect on the magnesium ion weakens, and the dwell time of magnesium ions within the channel decreases. Thus, the kinetics of the binding reaction between magnesium and the NMDAR channel are such that magnesium periodically unbinds and leaves the channel, only to be replaced by another magnesium ion. During the (very brief) time that the magnesium is absent from the open channel, other ions (such as sodium and calcium) can flow through the channel. However, when the cell is more hyperpolarized, the bound state of magnesium is stabilized and it leaves the channel less often and for a shorter period of time (on average). When the cell is less hyperpolarized, the magnesium leaves the channel more often and stays away for longer (on average). Hence, the magnesium blockade of the open NMDAR channel is membrane voltage-dependent.
While the NMDAR channel itself displays little or no voltage dependence (its open channel I/V curve is more or less linear), the voltage dependence of the magnesium block effectively, if indirectly, confers voltage dependence to this channel. Thus, in effect, the NMDAR channel is both a ligand-gated and voltage-gated channel at the same time. This fact is critical to the function of the NMDAR as a Hebbian coincidence detector. More strictly speaking, inward cationic current (sodium or calcium) through the open unblocked NMDAR does decrease with depolarization (because of the decreased electrochemical "driving force"), but the voltage-dependent unblocking seems to outweigh this decrease in driving force, so the calcium influx into the spine caused by a pair of appropriately timed pre- and postsynaptic spikes significantly exceeds the sum of the influxes due to the individual spikes alone. This extra, or "nonlinear", calcium entry triggers the strength change.