Rotations in 4-dimensional Euclidean space

In mathematics, the group of rotations about a fixed point in four-dimensional Euclidean space is denoted SO(4). The name comes from the fact that it is the special orthogonal group of order 4.

In this article rotation means rotational displacement. For the sake of uniqueness rotation angles are assumed to be in the segment [0, π] except where mentioned or clearly implied by the context otherwise.

A "fixed plane" is a plane for which every vector in the plane is unchanged after the rotation. An "invariant plane" is a plane for which every vector in the plane, although it may be affected by the rotation, remains in the plane after the rotation.

Four-dimensional rotations are of two types: simple rotations and double rotations.

A simple rotation R about a rotation centre O leaves an entire plane A through O (axis-plane) fixed. Every plane B that is completely orthogonal[a] to A intersects A in a certain point P. Each such point P is the centre of the 2D rotation induced by R in B. All these 2D rotations have the same rotation angle α.

Half-lines from O in the axis-plane A are not displaced; half-lines from O orthogonal to A are displaced through α; all other half-lines are displaced through an angle less than α.

For each rotation R of 4-space (fixing the origin), there is at least one pair of orthogonal 2-planes A and B each of which is invariant and whose direct sum AB is all of 4-space. Hence R operating on either of these planes produces an ordinary rotation of that plane. For almost all R (all of the 6-dimensional set of rotations except for a 3-dimensional subset), the rotation angles α in plane A and β in plane B – both assumed to be nonzero – are different. The unequal rotation angles α and β satisfying −π < α, β < π are almost[b] uniquely determined by R. Assuming that 4-space is oriented, then the orientations of the 2-planes A and B can be chosen consistent with this orientation in two ways. If the rotation angles are unequal (αβ), R is sometimes termed a "double rotation".

In that case of a double rotation, A and B are the only pair of invariant planes, and half-lines from the origin in A, B are displaced through α and β respectively, and half-lines from the origin not in A or B are displaced through angles strictly between α and β.

If the rotation angles of a double rotation are equal then there are infinitely many invariant planes instead of just two, and all half-lines from O are displaced through the same angle. Such rotations are called isoclinic or equiangular rotations, or Clifford displacements. Beware: not all planes through O are invariant under isoclinic rotations; only planes that are spanned by a half-line and the corresponding displaced half-line are invariant.

Assuming that a fixed orientation has been chosen for 4-dimensional space, isoclinic 4D rotations may be put into two categories. To see this, consider an isoclinic rotation R, and take an orientation-consistent ordered set OU, OX, OY, OZ of mutually perpendicular half-lines at O (denoted as OUXYZ) such that OU and OX span an invariant plane, and therefore OY and OZ also span an invariant plane. Now assume that only the rotation angle α is specified. Then there are in general four isoclinic rotations in planes OUX and OYZ with rotation angle α, depending on the rotation senses in OUX and OYZ.

We make the convention that the rotation senses from OU to OX and from OY to OZ are reckoned positive. Then we have the four rotations R1 = (+α, +α), R2 = (−α, −α), R3 = (+α, −α) and R4 = (−α, +α). R1 and R2 are each other's inverses; so are R3 and R4. As long as α lies between 0 and π, these four rotations will be distinct.

Isoclinic rotations with like signs are denoted as left-isoclinic; those with opposite signs as right-isoclinic. Left- and right-isoclinic rotations are represented respectively by left- and right-multiplication by unit quaternions; see the paragraph "Relation to quaternions" below.

The four rotations are pairwise different except if α = 0 or α = π. The angle α = 0 corresponds to the identity rotation; α = π corresponds to the central inversion, given by the negative of the identity matrix. These two elements of SO(4) are the only ones that are simultaneously left- and right-isoclinic.

Left- and right-isocliny defined as above seem to depend on which specific isoclinic rotation was selected. However, when another isoclinic rotation R′ with its own axes OU′, OX′, OY′, OZ′ is selected, then one can always choose the order of U′, X′, Y′, Z′ such that OUXYZ can be transformed into OU′X′Y′Z′ by a rotation rather than by a rotation-reflection (that is, so that the ordered basis OU′, OX′, OY′, OZ′ is also consistent with the same fixed choice of orientation as OU, OX, OY, OZ). Therefore, once one has selected an orientation (that is, a system OUXYZ of axes that is universally denoted as right-handed), one can determine the left or right character of a specific isoclinic rotation.

Each plane through the rotation centre O is the axis-plane of a commutative subgroup isomorphic to SO(2). All these subgroups are mutually conjugate in SO(4).

Each pair of completely orthogonal planes through O is the pair of invariant planes of a commutative subgroup of SO(4) isomorphic to SO(2) × SO(2).

These groups are maximal tori of SO(4), which are all mutually conjugate in SO(4). See also Clifford torus.

All left-isoclinic rotations form a noncommutative subgroup S3L of SO(4), which is isomorphic to the multiplicative group S3 of unit quaternions. All right-isoclinic rotations likewise form a subgroup S3R of SO(4) isomorphic to S3. Both S3L and S3R are maximal subgroups of SO(4).

Each left-isoclinic rotation commutes with each right-isoclinic rotation. This implies that there exists a direct product S3L × S3R with normal subgroups S3L and S3R; both of the corresponding factor groups are isomorphic to the other factor of the direct product, i.e. isomorphic to S3. (This is not SO(4) or a subgroup of it, because S3L and S3R are not disjoint: the identity I and the central inversion I each belong to both S3L and S3R.)

Each 4D rotation A is in two ways the product of left- and right-isoclinic rotations AL and AR. AL and AR are together determined up to the central inversion, i.e. when both AL and AR are multiplied by the central inversion their product is A again.

This implies that S3L × S3R is the universal covering group of SO(4) — its unique double cover — and that S3L and S3R are normal subgroups of SO(4). The identity rotation I and the central inversion I form a group C2 of order 2, which is the centre of SO(4) and of both S3L and S3R. The centre of a group is a normal subgroup of that group. The factor group of C2 in SO(4) is isomorphic to SO(3) × SO(3). The factor groups of S3L by C2 and of S3R by C2 are each isomorphic to SO(3). Similarly, the factor groups of SO(4) by S3L and of SO(4) by S3R are each isomorphic to SO(3).

The odd-dimensional rotation groups do not contain the central inversion and are simple groups.

SO(4) is different: there is no conjugation by any element of SO(4) that transforms left- and right-isoclinic rotations into each other. Reflections transform a left-isoclinic rotation into a right-isoclinic one by conjugation, and vice versa. This implies that under the group O(4) of all isometries with fixed point O the distinct subgroups S3L and S3R are conjugate to each other, and so cannot be normal subgroups of O(4). The 5D rotation group SO(5) and all higher rotation groups contain subgroups isomorphic to O(4). Like SO(4), all even-dimensional rotation groups contain isoclinic rotations. But unlike SO(4), in SO(6) and all higher even-dimensional rotation groups any two isoclinic rotations through the same angle are conjugate. The set of all isoclinic rotations is not even a subgroup of SO(2N), let alone a normal subgroup.

SO(4) is commonly identified with the group of orientation-preserving isometric linear mappings of a 4D vector space with inner product over the real numbers onto itself.

With respect to an orthonormal basis in such a space SO(4) is represented as the group of real 4th-order orthogonal matrices with determinant +1.

A 4D rotation given by its matrix is decomposed into a left-isoclinic and a right-isoclinic rotation as follows:

M has rank one and is of unit Euclidean norm as a 16D vector if and only if A is indeed a 4D rotation matrix[citation needed]. In this case there exist real numbers a, b, c, d and p, q, r, s such that

There are exactly two sets of a, b, c, d and p, q, r, s such that a2 + b2 + c2 + d2 = 1 and p2 + q2 + r2 + s2 = 1. They are each other's opposites.

The first factor in this decomposition represents a left-isoclinic rotation, the second factor a right-isoclinic rotation. The factors are determined up to the negative 4th-order identity matrix, i.e. the central inversion.

A point in 4-dimensional space with Cartesian coordinates (u, x, y, z) may be represented by a quaternion P = u + xi + yj + zk.

A left-isoclinic rotation is represented by left-multiplication by a unit quaternion QL = a + bi + cj + dk. In matrix-vector language this is

Likewise, a right-isoclinic rotation is represented by right-multiplication by a unit quaternion QR = p + qi + rj + sk, which is in matrix-vector form

In the preceding section (#Isoclinic decomposition) it is shown how a general 4D rotation is split into left- and right-isoclinic factors.

According to the German mathematician Felix Klein this formula was already known to Cayley in 1854[citation needed].

which shows that left-isoclinic and right-isoclinic rotations commute.

The four eigenvalues of a 4D rotation matrix generally occur as two conjugate pairs of complex numbers of unit magnitude. If an eigenvalue is real, it must be ±1, since a rotation leaves the magnitude of a vector unchanged. The conjugate of that eigenvalue is also unity, yielding a pair of eigenvectors which define a fixed plane, and so the rotation is simple. In quaternion notation, a proper (i.e., non-inverting) rotation in SO(4) is a proper simple rotation if and only if the real parts of the unit quaternions QL and QR are equal in magnitude and have the same sign.[c] If they are both zero, all eigenvalues of the rotation are unity, and the rotation is the null rotation. If the real parts of QL and QR are not equal then all eigenvalues are complex, and the rotation is a double rotation.

Our ordinary 3D space is conveniently treated as the subspace with coordinate system 0XYZ of the 4D space with coordinate system UXYZ. Its rotation group SO(3) is identified with the subgroup of SO(4) consisting of the matrices

In Van Elfrinkhof's formula in the preceding subsection this restriction to three dimensions leads to p = a, q = −b, r = −c, s = −d, or in quaternion representation: QR = QL′ = QL−1. The 3D rotation matrix then becomes

which is the representation of the 3D rotation by its Euler–Rodrigues parameters: a, b, c, d.

The corresponding quaternion formula P′ = QPQ−1, where Q = QL, or, in expanded form:

Rotations in 3D space are made mathematically much more tractable by the use of spherical coordinates. Any rotation in 3D can be characterized by a fixed axis of rotation and an invariant plane perpendicular to that axis. Without loss of generality, we can take the xy-plane as the invariant plane and the z-axis as the fixed axis. Since radial distances are not affected by rotation, we can characterize a rotation by its effect on the unit sphere (2-sphere) by spherical coordinates referred to the fixed axis and invariant plane:

Every rotation in 3D space has an invariant axis-line which is unchanged by the rotation. The rotation is completely specified by specifying the axis of rotation and the angle of rotation about that axis. Without loss of generality, this axis may be chosen as the z-axis of a Cartesian coordinate system, allowing a simpler visualization of the rotation.

Analogous to the 3D case, every rotation in 4D space has at least two invariant axis-planes which are left invariant by the rotation and are completely orthogonal (i.e. they intersect at a point). The rotation is completely specified by specifying the axis planes and the angles of rotation about them. Without loss of generality, these axis planes may be chosen to be the uz- and xy-planes of a Cartesian coordinate system, allowing a simpler visualization of the rotation.

Four-dimensional rotations can be derived from Rodrigues' rotation formula and the Cayley formula. Let A be a 4 × 4 skew-symmetric matrix. The skew-symmetric matrix A can be uniquely decomposed as

into two skew-symmetric matrices A1 and A2 satisfying the properties A1A2 = 0, A13 = −A1 and A23 = −A2, where θ1i and θ2i are the eigenvalues of A. Then, the 4D rotation matrices can be obtained from the skew-symmetric matrices A1 and A2 by Rodrigues' rotation formula and the Cayley formula.[5]

Let A be a 4 × 4 nonzero skew-symmetric matrix with the set of eigenvalues

Moreover, the skew-symmetric matrices A1 and A2 are uniquely obtained as

is a rotation matrix in E4, which is generated by Rodrigues' rotation formula, with the set of eigenvalues

is a rotation matrix in E4, which is generated by Cayley's rotation formula, such that the set of eigenvalues of R is,

The generating rotation matrix can be classified with respect to the values θ1 and θ2 as follows: