[Illustration: FIG. 10. The course of the light and of the two half-beams in Michelson's most famous experiment. The light is split at A, one half sent towards B and back, the other half to C and back. Compare with Fig. 7.] _Michelson Experiment._ Mr. Michelson, however, of the United States, invented a plan that looked as if it really would discriminate; and, after overcoming many difficulties, he carried it out. It is described in the _Philosophical Magazine_ for 1887. Michelson's famous experiment consists in looking for interference between two half-beams of light, of which one has been sent to and fro _across_ the line of ether drift, and the other has been sent to and fro _along_ the line of ether drift. A semi-transparent mirror set at 45° is employed to split the beam, and a pair of normal and ordinary mirrors, set perpendicular to the two half-beams, are employed to return them back whence they came, so that they can enter the eye through an observing telescope. It differs essentially from the interference kaleidoscope, Fig. 7, inasmuch as there is now no luminous path B C, and no contour enclosed by the light. Each half-beam goes to and fro on its own path, and these paths, instead of being coincident, are widely separate,--one North and South, for instance, and the other East and West. Under these conditions the bands are much more tremulous than they were in the arrangement of Fig. 7, and are subject to every kind of disturbance. The apparatus has to be excessively steady, and no fluctuation even of temperature must be permitted in the path of either beam. To secure this, the source, the mirrors, and the observing telescope, were all mounted upon a massive stone slab; and this was floated in a bath of mercury. The slab could then be slowly turned round, so that sometimes the path A B and sometimes the path A C lay approximately along or athwart the direction of the earth's motion in space. And inasmuch as the motion along would take a little longer than the motion across, though everything else was accurately the same, some shift of the interference bands might be expected as the slab rotated. But whereas in all the experiments previously described the effect looked for was a first-order effect, of magnitude one in ten or twenty thousand,--depending, that is to say, on the first power of the ratio of speed of earth to speed of light,--the effect now to be expected depends on the _square_ of that same ratio, and therefore cannot be greater, even in the most favourable circumstances, than 1 part in a hundred million. It is easy to realise therefore that it is an exceptionally difficult experiment, and that it required both skill and pertinacity to perform it successfully. That it is an exceptionally difficult experiment will be realised when I say that it would fail in conclusiveness unless one part in 400 millions could be clearly detected.