In what way is a "cheat" when using smooth shading on a sphere approximation?
I'm not understanding how it's a cheat to use smooth shading on this?
Other questions that are lead after this may be: Why does it look better than the Gouraud Shaded Model and why is this not necessary generally applicable to any surface?
Sphere's don't have edges like the ones in that picture. The "silhouette" should be a circle. This is because the sphere was rendered as triangles and rasterized. Triangles can only approximate a sphere.
Compare this to ray casting for instance. Where we can directly test the intersection between a sphere and ray thus no need to approximate a sphere with other primitives.
What smooth shading lets us do is interpolate values across the face of a triangle to cheat and make something appear smoother. In reality a sphere (a uv sphere) would look like this:
There are two common ways to make the above sphere look smooth without adding anymore geometry. Both of which involve linearly interpolating values across the face of a triangle. Like barycentric interpolation. With barycentric interpolation, given values at 3 points (on a triangle) and a point we want to know the value for we can calculate a mix between the 3 values at the corners such that it smoothly varies across the face of the triangle.
In rasterization there are two popular ways to smooth shade. The later being used more often.
Gouraud: We calculate the post-lighting (shaded) color at each vertex and interpolate the color across the face of the triangle.
Phong: We calculate the shaded color at every pixel across the triangle, interpolating things like normals across the triangle.
EDIT: The difference between gouraud and phong is that gouraud assumes the color and light intensity change linearly with the normal. Whereas phong does not. For a simple light intensity = dot(normal, lightDir) = cos(angle between normal and lightDir) we can see the relationship between intensity and the normal is nonlinear. This can be well illustrated with the below image (see the seams on the specular highlight. This is the consequence of our linear interpolation):
