Euler Angles

In July 1969, as Apollo 11 coasted toward the Moon, command-module pilot Michael Collins radioed Houston a request: "How about sending me a fourth gimbal for Christmas?" The spacecraft's guidance platform hung in three nested gimbal rings, its orientation tracked by three angles — and Collins knew that if the stack ever tilted into the wrong attitude, two of those rings would fold into alignment, the platform would lose a degree of freedom, and the crew would have to realign the entire inertial system by sextant and starlight. NASA had shaved off the fourth ring to save weight. The three angles that remained, with their power and their one fatal blind spot, are the Euler angles. The mathematical problem they solve is ancient and universal: a rigid body pivoting about a fixed point can point anywhere and be twisted anyhow, yet its orientation is captured by exactly three numbers. Euler's recipe builds any orientation from three successive rotations — a turn $\phi$ about the vertical, a tilt $\theta$ about a moving hinge line called the line of nodes, and a final spin $\psi$ about the body's own axis. Multiply the three elementary matrices together and you hold the complete dictionary between laboratory coordinates and body coordinates. For the mechanics of tops and gyroscopes these three angles are not merely coordinates — they are the motion: $\dot\phi$ is precession, $\dot\theta$ is nutation, $\dot\psi$ is spin. That direct physical meaning is why, three centuries on, every treatment of rigid-body dynamics from Goldstein to spacecraft flight software still begins exactly here.

Key takeaways

Euler angles $(\phi,\theta,\psi)$ are three successive rotation angles that map the fixed (space) frame onto the body-fixed frame, fully specifying a rigid body's orientation — the minimal chart on the three-dimensional rotation group $SO(3)$ .
In the standard $z$ - $x'$ - $z''$ ( $zxz$ ) convention: $\phi$ is precession about the space $z$ -axis, $\theta$ is nutation about the line of nodes, and $\psi$ is spin about the body symmetry axis.
The total rotation matrix is the ordered product $A = B(\psi)\,C(\theta)\,D(\phi)$ — the last rotation applied sits leftmost, and $A$ carries space components to body components.
$A$ is orthogonal with $\det A = +1$ ; its entries encode the angles directly ( $a_{33}=\cos\theta$ ), so the angles can be extracted from any attitude matrix.
At $\theta = 0,\pi$ the first and third rotation axes coincide and only $\phi+\psi$ is physically meaningful — gimbal lock, the unavoidable coordinate singularity of every three-angle chart.
Conventions differ across fields ( $zxz$ for tops, $zyx$ yaw–pitch–roll for aircraft, active vs. passive); state yours before writing a single component.
Key idea: any rigid-body orientation is reached by three successive rotations — $\phi$ about space $z$ , then $\theta$ about the line of nodes, then $\psi$ about body $z$ — so the orientation is fully captured by the ordered triple $(\phi,\theta,\psi)$ and the rotation matrix $A=B(\psi)C(\theta)D(\phi)$ .

Where this lives today

Every attitude-control system flying today — satellites, launch vehicles, drones, the phone in your pocket reporting yaw–pitch–roll — computes with the mathematics of this page, though usually one representation removed. Flight software integrates orientation as a quaternion (four numbers, no singularity, cheap to renormalize) and converts to Euler angles at the edges: for pilots' displays, pointing budgets, and human-readable telemetry. The conversion is exactly the extraction recipe (7) with its $1/\sin\theta$ conditioning, and aerospace engineers still speak of "gimbal-lock avoidance" even in gimbal-less strapdown systems, because the singularity lives in the chart, not the hardware. The same three angles organize modern molecular and materials physics. The rotational Hamiltonian of a polyatomic molecule is written in Euler angles relative to its principal axes, and its quantized spectrum — measured by microwave spectroscopy — is how radio astronomers identify molecules in interstellar clouds. In crystallography and materials science, "orientation maps" of the grains in a polycrystalline metal are literally images whose pixel values are Euler-angle triples (the Bunge $zxz'$ convention), and texture analysis integrates over $SO(3)$ with the volume element $\sin\theta\,\mathrm d\theta\,\mathrm d\phi\,\mathrm d\psi$ — the Jacobian factor that is one more echo of the $\sin\theta$ in equation (7). Robotics closes the loop back to Apollo: every six-axis arm has wrist singularities where two joint axes align, and motion planners route around them precisely the way Armstrong and Collins steered the command module around the red gimbal-lock warning zone on their eight-ball display.

The physical picture

Watch a spinning top on a table and you can see all three Euler angles at once. The tilt of its axis away from the vertical is $\theta$ . The slow sweep of that tilted axis around the vertical — the cone the axis traces — is $\phi$ advancing. And the blur of the top whirling about its own axis is $\psi$ . Euler's angles are not an arbitrary bookkeeping scheme; for a top they are the three motions your eye already separates: precession, nutation, spin.

Figure

Three turns build any orientation — precession, then tilt about the line of nodes, then spin

The hinge of the whole construction is the line of nodes: the intersection of the space $xy$ -plane with the body's equatorial ( $x'y'$ ) plane. It is where the first rotation parks the $x$ -axis, it is the axis about which the tilt $\theta$ happens, and it belongs to both planes at once — one foot in the laboratory, one foot in the body. Every trigonometric factor in every Euler-angle formula is ultimately this line being projected one way or the other.

Figure

The line of nodes — the hinge shared by the space plane and the body plane

The second image to hold onto is three nested gimbal rings — the physical hardware of Apollo's guidance platform. The outer ring turns by $\phi$ about a vertical axis, the middle ring tilts by $\theta$ , the inner ring spins by $\psi$ . When the middle ring lies flat ( $\theta = 0$ ), the outer and inner axes point along the same line: turning one is indistinguishable from turning the other, and the mechanism — like the mathematics — has lost a degree of freedom. That is gimbal lock, and it is not a defect of sloppy engineering but a theorem: no three-angle chart can cover the curved group $SO(3)$ without a singular point, for the same topological reason no flat map covers the whole globe without distortion. One more warning your intuition needs upfront: order matters absolutely. Rotate a book $90°$ about the vertical and then $90°$ about a horizontal axis; now repeat in the other order — it ends up oriented differently. Rotations do not commute, which is why the recipe must fix the sequence $\phi$ -then- $\theta$ -then- $\psi$ , and why $SO(3)$ is a genuinely non-abelian group rather than three independent dials.

Building the rotation, one angle at a time

The strategy is composition. Each of the three elementary rotations is trivial on its own — a $2\times 2$ rotation block padded with a 1 — and the full transformation is their ordered product. In the passive ( $zxz$ ) convention of Goldstein and Landau, the matrix acts on components: feed it the components of any fixed vector in space axes and it returns the components of the same vector in body axes:

A(\phi,\theta,\psi) = B(\psi)\,C(\theta)\,D(\phi) \qquad (1)

where: $D(\phi)$ — rotation by $\phi$ about the space $z$ -axis; $C(\theta)$ — rotation by $\theta$ about the line of nodes (the intermediate $x$ -axis); $B(\psi)$ — rotation by $\psi$ about the body $z$ -axis. Read the product right to left: $D$ acts first on the space components, then $C$ acts on what $D$ produced, then $B$ — the last rotation applied sits leftmost. Writing them in the order applied, left to right, is the single most common composition error. The first factor turns the horizontal plane, carrying the $x$ -axis into the line of nodes:

D(\phi)=\begin{pmatrix}\cos\phi & \sin\phi & 0\\ -\sin\phi & \cos\phi & 0\\ 0 & 0 & 1\end{pmatrix} \qquad (2)

where: $\phi \in [0,2\pi)$ — the precession angle. The third row and column are trivial because the rotation axis is $z$ itself: anything along $z$ is untouched. The second factor is the tilt — the only rotation that moves the $z$ -axis, and therefore the only place the angle $\theta$ between space $z$ and body $z$ can enter:

C(\theta)=\begin{pmatrix}1 & 0 & 0\\ 0 & \cos\theta & \sin\theta\\ 0 & -\sin\theta & \cos\theta\end{pmatrix} \qquad (3)

where: $\theta \in [0,\pi]$ — the nutation angle, measured about the line of nodes. The restricted range is deliberate: allowing $\theta$ beyond $\pi$ would describe orientations already reachable with $\theta' = 2\pi - \theta$ and shifted $\phi, \psi$ — the chart would double-cover. The third factor spins the body about its own (new) $z$ -axis and has the same form as the first:

B(\psi)=\begin{pmatrix}\cos\psi & \sin\psi & 0\\ -\sin\psi & \cos\psi & 0\\ 0 & 0 & 1\end{pmatrix} \qquad (4)

where: $\psi \in [0,2\pi)$ — the spin angle about the body symmetry axis. That $D$ and $B$ are matrices of identical form about different physical axes is exactly what makes $\theta \to 0$ dangerous: remove the tilt between them and they merge into one rotation.

Derivation AComposing the full Euler rotation matrix

We are deriving the explicit $3\times3$ matrix $A(\phi,\theta,\psi)$ , equation (5), by multiplying out the product $B(\psi)C(\theta)D(\phi)$ , and then verifying that the result is a proper rotation — orthogonal with determinant $+1$ . The route is brute-force but disciplined: do the inner product $C\,D$ first (it is the cheaper multiplication, since $C$ has a trivial first row), then apply $B$ from the left. The things to watch are the order (right-to-left, in the order the rotations are applied) and the bookkeeping of which frame each intermediate matrix refers to; the payoff is a matrix whose individual entries turn out to be readable — several of them are single trigonometric products that let you extract the angles back out of any numerical attitude matrix.

Step 1. Multiply the tilt into the precession. Using (2) and (3),

C(\theta)D(\phi) = \begin{pmatrix}\cos\phi & \sin\phi & 0\\ -\cos\theta\sin\phi & \cos\theta\cos\phi & \sin\theta\\ \sin\theta\sin\phi & -\sin\theta\cos\phi & \cos\theta\end{pmatrix}

Why the third row is already final: $B(\psi)$ has $(0,0,1)$ as its third row, so the last row of $CD$ passes through the final multiplication untouched — the body $z$ -components are fixed once the tilt is done, because the spin happens about that very axis.

Step 2. Apply the spin from the left, $A = B(\psi)\,[C(\theta)D(\phi)]$ , to obtain the full matrix:

A=\begin{pmatrix}\cos\psi\cos\phi-\cos\theta\sin\phi\sin\psi & \cos\psi\sin\phi+\cos\theta\cos\phi\sin\psi & \sin\psi\sin\theta\\ -\sin\psi\cos\phi-\cos\theta\sin\phi\cos\psi & -\sin\psi\sin\phi+\cos\theta\cos\phi\cos\psi & \cos\psi\sin\theta\\ \sin\theta\sin\phi & -\sin\theta\cos\phi & \cos\theta\end{pmatrix} \qquad (5)

where: rows are body axes, columns are space axes: $a_{ij} = \hat{\mathbf e}_i' \cdot \hat{\mathbf e}_j$ , the direction cosines between the two frames. Why several entries are simple while four are two-term sums: the simple entries involve the two $z$ -axes (each $z$ -axis is touched by only one or two of the three rotations), while the upper-left $2\times2$ block mixes all three angles — both equatorial planes have been turned twice.

Step 3. Verify orthogonality. Each factor satisfies $D^TD = C^TC = B^TB = \mathbb{1}$ by direct multiplication, so

A^TA = D^TC^TB^T\,BCD = \mathbb{1}, \qquad A^{-1} = A^T \qquad (6)

Why this had to happen: rotations preserve lengths, $|\mathbf r'|^2 = \mathbf r^TA^TA\,\mathbf r = |\mathbf r|^2$ for every $\mathbf r$ , which forces $A^TA = \mathbb 1$ . Orthogonality is not a happy accident of the algebra; it is the defining property of rotation.

Step 4. Verify the determinant. Each elementary matrix has $\det = \cos^2 + \sin^2 = +1$ , and the determinant is multiplicative, so $\det A = +1$ : $A$ is a proper rotation, never a reflection.

Step 5. Read the angles back out of the matrix. The third row and third column of (5) are single products, so given a numerical attitude matrix the angles follow immediately:

\theta = \arccos a_{33}, \qquad \phi = \operatorname{atan2}(a_{31},\,-a_{32}), \qquad \psi = \operatorname{atan2}(a_{13},\,a_{23}) \qquad (7)

where: $\operatorname{atan2}(y,x)$ — the two-argument arctangent that returns the correct quadrant. Why atan2 and not a ratio: $a_{31}/(-a_{32}) = \tan\phi$ loses the quadrant and blows up at $\cos\phi = 0$ ; keeping numerator and denominator separate preserves the full $[0,2\pi)$ range. The recipe fails only when $\sin\theta = 0$ , where $a_{31} = a_{32} = a_{13} = a_{23} = 0$ and only $\phi + \psi$ is defined — gimbal lock again. *Verified: the matrix (5) matches Goldstein 3rd ed. eq. (4.46); numerically, $A(30°,45°,60°)$ gives $a_{33} = \cos 45° = 0.7071$ , $A^TA = \mathbb 1$ and $\det A = 1.0000$ to machine precision.*

Go deeper · Derivation B · 2 minWhy exactly three angles — the dimension of the rotation group

Derivation BWhy exactly three angles — the dimension of the rotation group

The goal is the claim underneath the whole construction: rigid-body orientation has exactly three degrees of freedom, and any orientation is a single rotation about some axis (Euler's rotation theorem) — so a three-angle recipe like $zxz$ can, singularities aside, reach everything. The strategy is two independent arguments that meet in the middle: a constraint count on the entries of an orthogonal matrix gives the dimension, and an eigenvalue argument produces the fixed rotation axis. Watch the determinant condition — it is what excludes mirror reflections, which also preserve lengths but cannot be reached by any continuous rotation from rest.

Step 1. Count the constraints. An orientation is a $3\times3$ matrix $A$ with nine entries, constrained by orthonormality: the three rows must be unit vectors (3 conditions) and mutually perpendicular (3 conditions). That leaves

9 - 6 = 3

free parameters. Why this is the whole content of "three Euler angles": any non-degenerate triple of angles is just one choice of coordinates on this three-dimensional space; $zxz$ , $zyz$ , and yaw–pitch–roll are different charts on the same group manifold.

Step 2. Show a rotation axis always exists. The eigenvalues of a real orthogonal matrix have unit modulus, and complex ones come in conjugate pairs $e^{\pm i\alpha}$ . For a $3\times3$ matrix the eigenvalues are $\{e^{i\alpha}, e^{-i\alpha}, \lambda_3\}$ with $\lambda_3$ real, and

\det A = e^{i\alpha}e^{-i\alpha}\lambda_3 = \lambda_3 = +1

Why this is Euler's theorem: the eigenvector $\hat{\mathbf n}$ belonging to $\lambda_3 = 1$ satisfies $A\hat{\mathbf n} = \hat{\mathbf n}$ — a direction left completely unchanged. Every rigid displacement about a fixed point, however it was assembled, is equivalent to one rotation by some angle $\alpha$ about this axis.

Step 3. Connect the two pictures. The axis–angle pair $(\hat{\mathbf n}, \alpha)$ is also three numbers (two for the unit vector, one for the angle) — the same dimension counted in Step 1, as it must be. Euler angles trade the democratic axis–angle description for a hierarchical one adapted to a distinguished vertical (gravity, the lab $z$ ) and a distinguished body axis (the symmetry axis of the top) — which is exactly why $\phi, \theta, \psi$ acquire the individual physical meanings precession, tilt, spin. Verified: the trace of (5) gives $\operatorname{tr}A = 1 + 2\cos\alpha$ , and for $A(30°,45°,60°)$ this yields $\alpha = 86.9°$ — a single equivalent rotation, as the theorem demands; matches Goldstein §4.6.

The kinematic dividend

Why go to this trouble? Because the moment the angles exist, their rates become the natural language of rotational motion. Each angle changes about its own axis, and infinitesimal rotations — unlike finite ones — add as vectors, so the angular velocity of the body is simply

\boldsymbol{\omega} = \dot\phi\,\hat{\mathbf z} + \dot\theta\,\hat{\mathbf n} + \dot\psi\,\hat{\mathbf e}_3 \qquad (8)

where: $\hat{\mathbf z}$ — the space vertical; $\hat{\mathbf n}$ — the line of nodes; $\hat{\mathbf e}_3$ — the body symmetry axis. The three axes are not mutually orthogonal, so extracting usable components requires projecting this sum onto the body principal axes — that projection, the resulting formulas $\omega_1 = \dot\phi\sin\theta\sin\psi + \dot\theta\cos\psi$ and companions, and their inversion are the business of the sibling article Angular Velocity in Terms of Euler Angles, which owns the derivation. The dividend that matters for dynamics is the kinetic energy. Feeding the body-frame components into $T = \tfrac12\sum_i I_i\omega_i^2$ and specializing to a symmetric top ( $I_1 = I_2$ ) collapses all the $\psi$ -dependence:

T = \tfrac12 I_1\!\left(\dot\theta^2 + \dot\phi^2\sin^2\theta\right) + \tfrac12 I_3\!\left(\dot\psi + \dot\phi\cos\theta\right)^2 \qquad (9)

where: $I_1 = I_2$ — the transverse principal moments; $I_3$ — the moment about the symmetry axis. Look at what survives: $T$ depends on the angle $\theta$ but not on $\phi$ or $\psi$ themselves. Add a gravitational potential $V = Mg\ell\cos\theta$ — also $\phi$ - and $\psi$ -free — and both are cyclic coordinates with conserved momenta $p_\phi$ and $p_\psi$ . That single observation is what makes the heavy symmetric top exactly solvable, and it is invisible in any parameterization less adapted to the problem. Choosing Euler angles is not notation; it is half the solution.

Which equation answers which question

you want	use	notes
body components of a space vector	$\mathbf r_{\text{body}} = A\,\mathbf r_{\text{space}}$ , eq. (5)	passive $zxz$ convention
space components of a body vector	$\mathbf r_{\text{space}} = A^T\mathbf r_{\text{body}}$ , eq. (6)	rows of $A$ = body axes in space
angles from an attitude matrix	extraction recipe, eq. (7)	fails only at $\sin\theta = 0$
where the symmetry axis points	third row of $A$ : $(\sin\theta\sin\phi,\,-\sin\theta\cos\phi,\,\cos\theta)$	$\psi$ -independent, as it must be
angular velocity from angle rates	eq. (8), then project	derivation in the sibling article
top Lagrangian	eq. (9) $-\,Mg\ell\cos\theta$	$\phi,\psi$ cyclic — two conserved momenta

Limiting cases

Case	Result	Meaning
aligned frames, $\theta = 0$	only $\phi + \psi$ defined	body $z$ coincides with space $z$ ; first and third rotations share an axis — gimbal lock
anti-aligned, $\theta = \pi$	only $\phi - \psi$ defined	the second singularity of the $zxz$ chart; no three-angle chart avoids both
pure spin, $\dot\phi = \dot\theta = 0$	$\boldsymbol\omega = \dot\psi\,\hat{\mathbf e}_3$	the body turns only about its symmetry axis
pure precession, $\dot\theta = \dot\psi = 0$	$\boldsymbol\omega = \dot\phi\,\hat{\mathbf z}$	the tilted body revolves rigidly about the vertical
small tilt, $\theta \ll 1$	$A \approx$ rotation by $\phi+\psi$ about $z$	extraction of separate $\phi,\psi$ becomes ill-conditioned as $1/\sin\theta$

Common trapA rotation matrix is a rotation matrix — surely the convention barely matters

It matters totally, and convention mismatch is the leading cause of wrong rigid-body computations. The $zxz$ scheme here (Goldstein, Landau) differs from $zyz$ (Taylor, quantum angular momentum) and from aerospace yaw–pitch–roll $zyx$ , in which the same symbol $\theta$ is a different angle with a different singularity ( $\theta = \pm 90°$ instead of $0, \pi$ ). Worse, each scheme exists in active form (rotate the vector, keep the frame) and passive form (rotate the frame, keep the vector), which are transposes of each other — so even with the axes agreed, a silent active/passive mismatch flips the sign of every angle. Before writing a single component: state the axis sequence, state active or passive, and check one known case (e.g. $\phi = 90°, \theta = \psi = 0$ should send $\hat{\mathbf y}$ to $\pm\hat{\mathbf x}$ — the sign tells you which convention you are actually in).

Worked examples

1 · Build the matrix and point the axis (confidence builder)

A gyroscope's orientation is given by $\phi = 30°$ , $\theta = 45°$ , $\psi = 60°$ in the $zxz$ convention. Construct $A$ , verify it is a proper rotation, and find where the body symmetry axis points in the laboratory.

Step 1. Evaluate the entries of (5) with $\cos 30° = 0.8660$ , $\cos 45° = 0.7071$ , $\cos 60° = 0.5$ . For instance $a_{11} = \cos 60°\cos 30° - \cos 45°\sin 30°\sin 60° = 0.4330 - 0.3062 = 0.1268$ , and the full matrix comes out

A = \begin{pmatrix} 0.1268 & 0.7803 & 0.6124\\ -0.9268 & -0.1268 & 0.3536\\ 0.3536 & -0.6124 & 0.7071 \end{pmatrix}

Step 2. Check propriety: multiplying, $AA^T = \mathbb 1$ to four decimals in every entry, and expanding the determinant gives $\det A = \boxed{1.0000}$ — a proper rotation, no reflection.

Step 3. The symmetry axis $\hat{\mathbf e}_3$ has body components $(0,0,1)$ , so its space components are $A^T(0,0,1)^T$ — the third row of $A$ :

\hat{\mathbf e}_3\big|_{\text{space}} = (\sin\theta\sin\phi,\ -\sin\theta\cos\phi,\ \cos\theta) = \boxed{(0.3536,\ -0.6124,\ 0.7071)}

Step 4. Sanity-read the answer: the axis is tilted $45°$ from vertical ( $\cos\theta = 0.7071$ ✓), and its horizontal shadow of length $\sin 45° = 0.7071$ points at azimuth set by $\phi = 30°$ alone — the spin $\psi$ never appears, because spinning a top about its own axis cannot move that axis.

2 · Recover the attitude from a star tracker (exam level)

A spacecraft star tracker delivers the attitude matrix

A = \begin{pmatrix} 0.5584 & 0.7444 & 0.3660\\ -0.6150 & 0.0755 & 0.7849\\ 0.5567 & -0.6634 & 0.5000 \end{pmatrix}

Extract the $zxz$ Euler angles, verify by recomposition, and estimate how the extraction degrades if the same tracker noise occurs near $\theta \approx 1°$ .

Step 1. Tilt first, from the corner entry: $\theta = \arccos a_{33} = \arccos 0.5000 = \boxed{60.0°}$ . Since $\sin\theta = 0.866$ is comfortably large, the remaining extractions are well-conditioned.

Step 2. Precession from the third row: $\phi = \operatorname{atan2}(a_{31}, -a_{32}) = \operatorname{atan2}(0.5567,\ 0.6634) = \boxed{40.0°}$ .

Step 3. Spin from the third column: $\psi = \operatorname{atan2}(a_{13}, a_{23}) = \operatorname{atan2}(0.3660,\ 0.7849) = \boxed{25.0°}$ .

Step 4. Recompose: plugging $(40°, 60°, 25°)$ back into (5) reproduces every entry of the given matrix to four decimals — the extraction is exact, not approximate.

Step 5. Now the conditioning question. The angle-bearing entries $a_{31}, a_{32}, a_{13}, a_{23}$ all carry a factor $\sin\theta$ , so a fixed matrix-entry error $\varepsilon$ produces angle errors of order $\varepsilon/\sin\theta$ . At $\theta = 60°$ the amplification is $1/0.866 = 1.15$ — harmless. At $\theta = 1°$ it is $1/\sin 1° = \boxed{57.3}$ : a modest $0.1\%$ tracker error becomes a $3°$ -scale error in $\phi$ and $\psi$ individually (their sum stays well-determined). This is gimbal lock as an error amplifier — the reason attitude software carries quaternions internally and converts to Euler angles only for human display.

Origins · 1 minLeonhard Euler and how this was discovered

Leonhard Euler

1707–1783 · St. Petersburg & Berlin academies

Euler is the most prolific mathematician who ever lived — roughly 850 works, a third of the entire mathematical output of the eighteenth century — and rigid-body motion is one of the fields he built almost single-handedly. Between his treatise on the rotation of solid bodies (1758-1765) he introduced the angles now bearing his name, derived the equations of motion for a spinning body, and proved the rotation theorem guaranteeing every reorientation is a single turn about some axis. The astonishing part is the timeline of his blindness: he lost his right eye in 1738 and nearly all remaining sight by 1766, yet his productivity increased — he computed in his head and dictated to assistants, producing on average one paper a week in his final decade. The notation you used on this page is his too: $e$ , $i$ , $\pi$ as standard symbols, $f(x)$ , the $\Sigma$ sign. Laplace's advice to students remains the field's consensus: "Read Euler, read Euler, he is the master of us all."

History

1758

— Euler's memoir on the rotation of solid bodies about a variable axis founds rigid-body kinematics, introducing what become the Euler angles.

1775

— Euler proves the rotation theorem: any displacement of a body about a fixed point is a single rotation about some axis.

1843

— Hamilton, walking past Broom Bridge in Dublin, discovers quaternions — the four-parameter, singularity-free alternative to three angles.

1897

— Klein and Sommerfeld begin the four-volume Theory of the Top, the definitive treatment of rigid-body motion in Euler angles.

1969

— Apollo 11's gimballed guidance platform flies with only three gimbals; the crew steers around gimbal lock, and Collins asks Houston for a fourth gimbal for Christmas.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics 3rd ed., §4.4 (The Euler Angles) · Landau & Lifshitz, Mechanics 3rd ed., §35 · Morin, Introduction to Classical Mechanics, Ch. 9 · Taylor, Classical Mechanics, Ch. 10

Provenance. Conventions and matrices follow Goldstein §4.4 ( $zxz$ , passive); a verifier should reproduce $D,C,B$ and the products, confirm $A^TA=\mathbb1$ and $\det A=+1$ , and recompute the body-frame $\omega_i$ (Goldstein Eq. 4.87 / Landau §35). The symmetric-top kinetic energy matches Goldstein Eq. 5.52. No empirical numbers are quoted. Cross-check Morin Ch. 9 and Taylor Ch. 10 for the same results in possibly transposed (active) form. Worked-example matrices on this page ( $A(30°,45°,60°)$ and the extraction $(40°,60°,25°)$ with conditioning factor $1/\sin 1° = 57.3$ ) are illustrative arithmetic recomputed from equations (5) and (7).

Builds on: Principal Axes and Principal Moments of Inertia · Moment of Inertia Tensor · Angular Momentum of a Rigid Body. Leads to: Angular Velocity in Terms of Euler Angles · Euler's Equations of Motion · Free Symmetric Top and Body-Cone/Space-Cone Precession · Classification of Rigid Bodies: Spherical, Symmetric, and Asymmetric Tops.

Why does orientation need exactly three numbers and not, say, two or four?

Count constraints on the direction-cosine matrix: nine entries, minus three unit-length conditions on the rows, minus three orthogonality conditions between them, leaves three free parameters — $SO(3)$ is a three-dimensional group. You can also count physically: two angles aim the symmetry axis (like latitude and longitude aim a telescope), and one more spins the body about that aimed axis. Four numbers are used in practice — quaternions — but the fourth is redundant by the unit-norm constraint; it buys freedom from singularities, not extra physical information. Fewer than three can never work: aiming the axis alone (two angles) says nothing about the twist around it.

How do I tell whether my textbook uses the active or the passive convention?

Apply the book's matrix to a trivial test case and see which way things move. Take $\phi = 90°$ , $\theta = \psi = 0$ , so $A = D(90°)$ , and act on the vector $(1,0,0)$ . In the passive convention (this page, Goldstein): the frame rotated $+90°$ , so the fixed vector's new components are $(0,-1,0)$ — the matrix has $\sin\phi$ in the upper right. In the active convention the vector rotates to $(0,+1,0)$ — $-\sin\phi$ upper right. The two conventions are exact transposes, so every formula survives the translation, but mixing them silently flips the sign of all three angles. Check once per book, write the result in the margin, and never trust memory.

Why is the middle rotation about the x-axis rather than z again — could I use a zzz sequence?

No — successive rotations about the same axis just add their angles: $D(\alpha)D(\beta) = D(\alpha+\beta)$ , so a $zzz$ sequence reaches only the one-parameter family of rotations about $z$ , not all of $SO(3)$ . The recipe needs the middle axis to be different so it can tip the $z$ -axis away from itself; any sequence in which consecutive axes differ works ( $zxz$ , $zyz$ , $xyz$ , …), and there are exactly twelve such sequences. What the $zxz$ family (first and third axes the same) buys you is symmetry adapted to problems with a distinguished vertical: $\phi$ and $\psi$ become rotations about the space vertical and the body axis respectively — precession and spin — which is why tops are always done this way.

Is gimbal lock a physical phenomenon or just a notation artifact?

Both, depending on what is built. For the mathematics, it is purely a chart singularity: the body at $\theta = 0$ is in a perfectly ordinary orientation and can rotate freely — only the labels $(\phi, \psi)$ degenerate, exactly as longitude degenerates at the North Pole while the pole itself is unremarkable. But if you build hardware whose axes physically realize the chart — a three-ring gimbal platform — the singularity becomes mechanical: with the rings coplanar there is no ring able to respond to torque about one direction, and the platform can be dragged off its inertial alignment. That is what Apollo crews genuinely had to avoid. Strapdown systems with quaternion software have no such hardware, yet the $1/\sin\theta$ blow-up still appears the moment anyone converts to Euler angles for display.

Why do aircraft use yaw–pitch–roll instead of the zxz angles used for tops?

Because you get to choose where to park the singularity, and each field parks it in the configuration it never visits. The $zxz$ chart is singular at $\theta = 0$ — body axis vertical — which for an aircraft is straight-and-level flight, the most common attitude; that would be a disaster. Tait–Bryan $zyx$ (yaw–pitch–roll) moves the singularity to pitch $= \pm 90°$ , pointing straight up or down, which airliners never do. A top, conversely, spends its life tilted at some working $\theta$ well away from $0$ , and $zxz$ rewards it with angles that are literally precession, nutation and spin, plus the cyclic-coordinate structure that solves the Lagrangian. Same group, different charts, chosen so the hole in the map falls over territory you do not visit.

If quaternions have no singularity, why teach Euler angles at all?

Because the two representations answer different needs. Quaternions win at computation: four numbers, no trigonometry in the update equation, no singular configuration, trivially renormalized against drift — which is why every flight computer and game engine integrates attitude as a quaternion. Euler angles win at meaning: they are directly measurable ( $\theta$ is the tilt you see), directly plottable, and — decisively for mechanics — they are generalized coordinates. You cannot write a Lagrangian in quaternion components without dragging in a constraint; in Euler angles $L = T - V$ is unconstrained, $\phi$ and $\psi$ are cyclic for the symmetric top, and two conservation laws fall out by inspection. The professional pattern is: think and derive in Euler angles, integrate in quaternions, convert at the interfaces.

How do I safely extract Euler angles from a rotation matrix near the singular configuration?

First detect it: compute $\sin\theta = \sqrt{a_{13}^2 + a_{23}^2}$ (equivalently $\sqrt{1 - a_{33}^2}$ ). If it is comfortably large, use the standard recipe $\theta = \arccos a_{33}$ , $\phi = \operatorname{atan2}(a_{31}, -a_{32})$ , $\psi = \operatorname{atan2}(a_{13}, a_{23})$ . If $\sin\theta$ is below a threshold (say $10^{-6}$ ), the matrix (5) collapses to a single rotation by $\phi \pm \psi$ about $z$ : recover that combination from the upper-left block, e.g. $\phi + \psi = \operatorname{atan2}(a_{12}, a_{11})$ at $\theta \approx 0$ , and set one angle by convention (commonly $\psi = 0$ ). Report the conditioning honestly: near the singularity the individual angles are undetermined by the physics, not merely by your algorithm — no amount of numerical care can recover information the chart does not contain.