The way in which a deformation of the atmosphere might explain the phenomenon is best seen by a diagram. First, it must be remarked that rays of light are only bent by the earth's atmosphere, or "refracted," if they enter it obliquely. If the atmosphere were of the same density throughout, like a piece of glass, then a vertical ray of light, A B (see Fig. 3), entering the atmosphere at B would suffer no bending or refraction, and a star shining from the direction A B would be seen truly in that direction from C. But an oblique ray, D E, would be bent on entering the atmosphere at E along the path EF, and a star shining along D E would appear from F to be shining along the dotted line G E F. The atmosphere is not of the same density throughout, but thins out as we go upwards from the earth; and in consequence there is no clear-cut surface, B E, and no sudden bending of the rays as at E: they are gradually bent at an infinite succession of imaginary surfaces. But it still remains true that there is no bending at all for vertical rays; and of oblique rays those most oblique are most bent. [Illustration: FIG. 4.] Now, suppose the atmosphere of the earth took up, owing to its revolution round the sun, an elongated shape like that indicated in diagram 4, and suppose the star to be at a great distance away to the right of the diagram. When the earth is in the position labelled "June," the light would fall vertically on the nose of the atmosphere at A, and there would be no refraction. Similarly in "December" the light would fall at C on the stern, also vertically, and there would be no refraction. [The rays from the distant star in December are to be taken as sensibly parallel to those received in June, notwithstanding that the earth is on the opposite side of the sun, as was remarked on p. 98.] But in March and September the rays would strike obliquely on the sides of the supposed figure, and thus be bent in opposite directions, as indicated by the dotted lines; and the extreme positions would thus occur in March and September, as had been observed. The explanation thus far seems satisfactory enough. But we have assumed the star to lie in the plane of the earth's orbit; and the stars under observation by Bradley did not lie in this plane, nor did they lie in directions equally inclined to it. Making the proper allowance for their directions, it was found impossible to fit in the facts with this hypothesis, which had ultimately to be abandoned. [Sidenote: Delay in finding real explanation.] [Sidenote: Bradley sets up another instrument at Wansted.] [Sidenote: Finds the right clue.] [Sidenote: A wind-vane on a boat.]