In this article we shall consider the evidence for reversals of the Earth's magnetic field.
When igneous rocks are formed, as the temperature of the rock falls beneath what is known as the Curie temperature (roughly speaking, the temperature above which a material cannot be magnetized and below which it can) iron-based minerals such as magnetite and hematite are magnetized by the Earth's magnetic field, indicating the directions of the Earth's north and south magnetic poles at the time that the rocks are formed.
Sedimentary rocks can also indicate these directions: when sediment is deposited gently in a low-energy environment such as the deep ocean floor, magnetized grains of magnetite and hematite will orient themselves to the Earth's magnetic field like so many tiny compass needles, indicating the directions of the north and south magnetic poles at the time the sediment was deposited.
When geologists realized that this was the case, it was immediately obvious that studying the magnetism of ancient rocks would tell them about the Earth's magnetic field as it existed in the deep past: paleomagnetism.
The study of paleomagnetism led to the discovery of magnetic field reversals. At present the Earth's magnetic field exhibits what is known as normal polarity: that is, it has the magnetic north pole near the geographic north pole, and the magnetic south pole near the geographic south pole. But studies of the paleomagnetism of ancient rocks showed evidence that in the past the Earth has sometimes had reversed polarity, with the magnetic north pole in the southern hemisphere and vice versa.
Apparently, then, the Earth periodically undergoes geomagnetic reversals, in which the north and south magnetic poles switch ends.
Geomagnetic reversals; how do we know?Edit
We can see today that magnetic minerals in sedimentary and igneous rocks align themselves with the present direction of the magnetic field; and the physics of this is well-understood — this is just what they ought to do.
We can also see that some rocks laid down in the past are magnetically aligned in the opposite direction. Moreover, all the rocks dated to a particular time in the past will have the same alignment: rocks that date to 60 million years ago will all have reversed polarity no matter where you look.
This really leaves us only with one plausible explanation. We can, to be sure, think of alternative implausible explanations. Perhaps the laws of physics themselves keep changing back and forth; or perhaps every now and then all the continents rapidly rotate 180 degrees in perfect synchrony. But it requires less of a stretch of the imagination to suppose that the poles themselves are moving.
This is supported by the fact that the poles are detectably moving. No-one has ever witnessed a geomagnetic reversal, which is hardly surprising if, as the evidence shows, they only take place about once every 100,000 years. but the poles certainly move: the difference between true north and magnetic north has been the subject of measurement and inquiry since the sixteenth century. At the present time of writing, the north magnetic pole is moving at around 40km/year, an unusually fast rate for anything to happen in geology. While the observation of this secular variation, as it is called, does not absolutely confirm the occurrence of geomagnetic reversals in the past, it certainly gives credence to the possibility of their occurrence.
Finally, in recent decades it has been possible to perform numerical simulations of the behavior of the Earth's core, starting with the work of Glatzmaier and Roberts in 1995 (which you can read here). Briefly, they put into a large and powerful computer the relevant laws of physics and the best estimates of the major forces and energies in the Earth's core. We may omit the details, but the important thing to note is that their model contained nothing explicitly relating to the motion of the poles, so that if they are found to move in the model, this must be implicit in the physics of the situation.
And their models did indeed exhibit polar reversals. The models also exhibited certain other features which must give us confidence in their broad accuracy: first, the fact that they exhibited a self-sustaining electromagnetic field at all; secondly, that they exhibit the sort of secular variation that we can observe today; third, that they exhibited one phenomenon which up until that time had never even been guessed at: that the solid inner core is rotating slightly faster than the rest of the planet: that is, it revolves slightly more than once a day. This really remarkable prediction has since been supported by seismological data (which you can read about here).
We may then regard the existence of geomagnetic reversals in the past as being on a very firm footing, since they should happen in principle and all the evidence shows that they did happen in practice.