Rotations of finite order are dense in SO(3)
Statement
Show that the rotations of whose angle is a rational multiple of (equivalently, the elements of finite order of that are rotations about an axis by such an angle) are dense in .
Hint
Every is a rotation about some axis. Keep the axis, and change the angle slightly to a rational multiple of .
Show solutionHide solution
Solution
Idea. An element of is determined by an axis and an angle. The axis is fixed during the approximation, and rational multiples of are dense in the set of angles.
Step 1: Euler's theorem (every has a fixed axis). The characteristic polynomial of is real of degree . Its roots have modulus (because preserves the Euclidean norm, and for an eigenvector with respect to the Hermitian norm, which also preserves). Non-real roots come in conjugate pairs , so either all three roots are real, equal to , or there is one real root and a pair . The product of the roots is .
- All real: the product forces or up to order; in both cases is a root.
- One real root and a pair: .
So is always an eigenvalue: there is a unit vector with .
Step 2: normal form. preserves (if then ). Choose an orthonormal basis of such that is positively oriented, and let . In this basis is with and , so . That is,
Step 3: approximation. Let with and (possible since is dense in ). Define
It is the rotation about the same axis by an angle in , and .
Because is orthogonal, the operator norm is invariant under conjugation by and
(Indeed, is the matrix of multiplication by the complex number on , whose operator norm is .)
Check (numerical). For a random , rounding the angle to a multiple of gives an error below in operator norm, as predicted.
Remark. The set of finite-order rotations is uncountable (different axes) and each has finite order, but the *group* of all rotations by rational angles about one fixed axis is countable. This is typical of density arguments: a dense set is topologically large but not necessarily "large" in cardinality.
Lesson. Reduce to a normal form with continuous parameters (here the angle), then approximate the parameters inside the normal form.