Angular Velocity in Terms of Euler Angles

Every drone that holds itself level in a gust is solving, hundreds of times per second, the problem this article solves once and for all. Its rate gyros report three numbers — the components of the angular velocity $\boldsymbol\omega$ along axes bolted to the airframe. Its autopilot needs three different numbers — how fast the orientation angles are changing. The two triples describe the same physical rotation, yet they live on different axes, and the dictionary between them is neither obvious nor, in one notorious configuration, even invertible. The dictionary exists because of a minor miracle of kinematics. A rigid body wearing Euler angles $(\phi,\theta,\psi)$ is doing three things at once: precessing at $\dot\phi$ about the space vertical, nodding at $\dot\theta$ about the line of nodes, spinning at $\dot\psi$ about its own axis. Finite rotations refuse to add — perform them in a different order and you get a different orientation — but infinitesimal rotations add like arrows. So the body's one true angular velocity is simply the vector sum $\boldsymbol\omega = \dot\phi\,\hat{\mathbf z} + \dot\theta\,\hat{\mathbf n} + \dot\psi\,\hat{\mathbf e}_3'$ , three contributions along three axes that are not mutually perpendicular and not fixed in any one frame. Projecting that sum onto the body's principal axes produces the working formulas of rigid-body dynamics — the components that feed the kinetic energy, the angular momentum, and Euler's equations. The same projection, run backwards, is what the drone computes; and the $1/\sin\theta$ that appears in the inverse is gimbal lock announcing itself in the rate equations. Master this one decomposition and the spinning top, the gyroscope, and the tumbling satellite all open with the same key.

Key takeaways

The Euler angles $(\phi,\theta,\psi)$ describe orientation through three successive rotations; their time derivatives $\dot\phi,\dot\theta,\dot\psi$ are angular velocities about three distinct, non-orthogonal axes.
The total angular velocity is the vector sum $\boldsymbol\omega=\dot\phi\,\hat{\mathbf z}+\dot\theta\,\hat{\mathbf n}+\dot\psi\,\hat{\mathbf e}_3'$ — exact and convention-free, because infinitesimal rotations add as vectors even though finite ones do not.
Projected onto the body principal axes ( $zxz$ convention): $\omega_1=\dot\phi\sin\theta\sin\psi+\dot\theta\cos\psi$ , $\omega_2=\dot\phi\sin\theta\cos\psi-\dot\theta\sin\psi$ , $\omega_3=\dot\phi\cos\theta+\dot\psi$ .
These components feed directly into $T=\tfrac12 I_i\omega_i^2$ and Euler's equations, making Euler angles the natural generalized coordinates for the Lagrangian mechanics of tops.
The rate matrix has determinant $-\sin\theta$ : at $\theta = 0,\pi$ the map $(\dot\phi,\dot\theta,\dot\psi)\to\boldsymbol\omega$ loses rank and its inverse blows up as $1/\sin\theta$ — gimbal lock in kinematic form.
For the symmetric top the projection delivers two cyclic coordinates and two conserved momenta, reducing the whole problem to one dimension.
Key idea: infinitesimal rotations are vectors, so the body's angular velocity is simply the vector sum $\boldsymbol\omega=\dot\phi\,\hat{\mathbf z}+\dot\theta\,\hat{\mathbf n}+\dot\psi\,\hat{\mathbf e}_3'$ of three rotations about three different axes — project this sum onto whatever basis your dynamics demands.

Where this lives today

The inverse map (6) — with its $1/\sin\theta$ — runs today inside essentially every object that knows its own orientation. A modern strapdown inertial navigation system straps three MEMS or fiber-optic rate gyros to the vehicle body: what they measure is precisely $(\omega_1, \omega_2, \omega_3)$ , sampled kilohertz-fast. Attitude is then propagated not through Euler rates but through the singularity-free quaternion kinematics $\dot q = \tfrac12 q \otimes \boldsymbol\omega$ — a direct engineering response to the determinant $-\sin\theta$ of equation (5) — and converted to yaw–pitch–roll only for pilots, operators, and telemetry. The Euler-rate equations survive in the conversion layer and in analysis, where their physical transparency is unmatched. The same decomposition organizes systems far from aerospace. Spacecraft attitude control laws for spin-stabilized satellites are written directly in $(\dot\phi, \dot\theta, \dot\psi)$ because steady precession is the nominal motion. In molecular physics, the symmetric-top kinetic energy (9) is quantized to give the rotational spectrum $E_{JK} = BJ(J+1) + (A-B)K^2$ , where $K$ — the quantum number of $p_\psi = I_3\omega_3$ — labels exactly the conserved axial component of equation (4); microwave spectroscopy of these levels is how radio astronomers fingerprint ammonia and methanol in interstellar clouds. And in biomechanics and motion capture, joint angular velocities extracted from marker data are Euler-rate reconstructions, complete with the same near-singular blow-ups when a joint approaches its chart's degenerate pose — a gimbal-lock artifact familiar to every animator who has watched a rigged shoulder snap through impossible orientations.

The physical picture

Watch a toy top in its last seconds before falling. It whirls about its own axis — that is $\dot\psi$ , the spin, a rotation about the body axis $\hat{\mathbf e}_3'$ . The tilted axis itself sweeps a slow cone around the vertical — that is $\dot\phi$ , the precession, a rotation about the space axis $\hat{\mathbf z}$ . And superposed on the sweep, the tilt angle bobs up and down — that is $\dot\theta$ , the nutation, a rotation about the line of nodes $\hat{\mathbf n}$ , the hinge lying in both the tabletop plane and the top's equatorial plane. Three visible motions, three axes, one body — and one angular velocity, obtained by gluing the three rotational arrows head to tail.

Figure

A top does three rotations at once — and its angular velocity is their vector sum

Why are the three axes so awkwardly chosen — one in the lab, one in the body, one straddling both? Because each is the axis its angle actually rotates about, and that is what makes the rates physically meaningful and the Lagrangian tractable. The price is that $(\dot\phi, \dot\theta, \dot\psi)$ are not the Cartesian components of $\boldsymbol\omega$ — they are oblique contributions that must be projected onto an orthonormal basis before any dynamics can be done. The basis of choice is the body's principal axes, where the inertia tensor is diagonal and $T = \tfrac12(I_1\omega_1^2 + I_2\omega_2^2 + I_3\omega_3^2)$ — the diagonalization owned by the sibling article on principal axes. The projection is pure trigonometry, and one picture organizes all of it: the space vertical $\hat{\mathbf z}$ seen from inside the body. It leans at angle $\theta$ from the body axis, so it donates $\cos\theta$ along $\hat{\mathbf e}_3'$ and a shadow of length $\sin\theta$ onto the equatorial plane, which the spin phase $\psi$ then splits between $\hat{\mathbf e}_1'$ and $\hat{\mathbf e}_2'$ :

Figure

The space vertical resolved in the body frame — the one projection behind all the formulas

Near $\theta = 0$ the picture degenerates: the vertical and the body axis coincide, precession and spin turn about the same line, and no measurement of $\boldsymbol\omega$ can tell $\dot\phi$ from $\dot\psi$ — only their sum survives. That is gimbal lock read off from kinematics alone, and it will reappear below as a determinant hitting zero.

The decomposition and its projections

Start from the statement that deserves to be called the master equation, because it is exact, convention-free, and the source of everything else on this page. During $\mathrm dt$ the body performs three infinitesimal rotations, and their angular velocities superpose:

\boldsymbol\omega=\dot\phi\,\hat{\mathbf z}+\dot\theta\,\hat{\mathbf n}+\dot\psi\,\hat{\mathbf e}_3' \qquad (1)

where: $\dot\phi$ — precession rate about the space vertical $\hat{\mathbf z}$ ; $\dot\theta$ — nutation rate about the line of nodes $\hat{\mathbf n}$ (the intersection of the space $xy$ -plane with the body equatorial plane); $\dot\psi$ — spin rate about the body symmetry axis $\hat{\mathbf e}_3'$ . Nothing here assumes a convention; only the components below do. The legitimacy of adding rotation rates as vectors is not obvious — finite rotations famously do not commute — and it deserves its own derivation.

Derivation AWhy angular velocities add although finite rotations do not commute

We are justifying the vector sum (1): three simultaneous rotation rates about three different axes combine into a single angular velocity vector by plain addition. The strategy is to expand each rotation to first order in the small angle, compose them, and watch the non-commuting parts die at the order we keep — the commutator of two rotations is second order in the angles, which is precisely why the paradox of non-commutativity evaporates for rates. The step to watch is the truncation: keeping the argument honest means saying exactly which terms are dropped and why they cannot survive the limit $\mathrm dt \to 0$ .

Step 1. Write an infinitesimal rotation as identity plus generator. A rotation by small angle $\mathrm d\alpha$ about unit axis $\hat{\mathbf a}$ acts on vectors as

R(\mathrm d\alpha,\hat{\mathbf a})\,\mathbf r \approx \mathbf r + \mathrm d\alpha\,(\hat{\mathbf a}\times\mathbf r) = \big(\mathbb 1+\mathrm d\alpha\,[\hat{\mathbf a}]_\times\big)\mathbf r

Why: the displacement of the tip of $\mathbf r$ under a small turn is tangent to the circle it rides, with magnitude $\mathrm d\alpha$ times the circle radius — exactly the cross product; $[\hat{\mathbf a}]_\times$ is the matrix that implements it.

Step 2. Compose the three Euler-rate rotations acting during the same $\mathrm dt$ :

R \approx \big(\mathbb 1+\mathrm d\psi[\hat{\mathbf e}_3']_\times\big)\big(\mathbb 1+\mathrm d\theta[\hat{\mathbf n}]_\times\big)\big(\mathbb 1+\mathrm d\phi[\hat{\mathbf z}]_\times\big) = \mathbb 1+\mathrm d\phi[\hat{\mathbf z}]_\times+\mathrm d\theta[\hat{\mathbf n}]_\times+\mathrm d\psi[\hat{\mathbf e}_3']_\times+\mathcal O(\mathrm d^2)

Why the order stops mattering: every product of two generators carries two small angles — $\mathrm d\phi\,\mathrm d\theta$ and the like. The entire non-commutativity of rotations lives in those crossed terms (this is the leading commutator in the Baker–Campbell–Hausdorff expansion), and they are second order.

Step 3. Divide by $\mathrm dt$ and take the limit. The angular velocity generator is

[\boldsymbol\omega]_\times=\lim_{\mathrm dt\to0}\frac{R-\mathbb 1}{\mathrm dt} = \dot\phi[\hat{\mathbf z}]_\times+\dot\theta[\hat{\mathbf n}]_\times+\dot\psi[\hat{\mathbf e}_3']_\times

and since $[\cdot]_\times$ is linear and invertible on vectors, the generators' sum is the vector sum (1). Why the dropped terms cannot return: they scale as $\mathrm dt^2/\mathrm dt = \mathrm dt \to 0$ . Non-commutativity is real — it is why the finite orientation depends on path — but it is a second-order effect invisible in the instantaneous rate. *Verified: a numerical check makes the orders concrete — composing two $90°$ rotations about $x$ then $z$ versus $z$ then $x$ leaves final orientations differing by a $120°$ rotation, while for two $1°$ rotations the discrepancy is only $0.0175°$ — quadratically small, exactly as Step 2 predicts.*

To do dynamics we need components on one orthonormal basis. Resolving the three oblique unit vectors of (1) onto the body principal axes (the projection drawn in the second figure) gives the workhorse formulas of the $zxz$ convention:

\omega_1=\dot\phi\sin\theta\sin\psi+\dot\theta\cos\psi \qquad (2)

where: $\omega_1$ — component along the body axis $\hat{\mathbf e}_1'$ . The first term is the equatorial shadow of the precession, phase-split by the spin angle; the second is the nutation, which lives along the line of nodes and therefore appears with the complementary phase factor.

\omega_2=\dot\phi\sin\theta\cos\psi-\dot\theta\sin\psi \qquad (3)

where: $\omega_2$ — component along $\hat{\mathbf e}_2'$ . Note the relative minus sign: as the body spins, the fixed nutation axis appears to rotate backwards in the body, so its components trade places with alternating sign as $\psi$ advances.

\omega_3=\dot\phi\cos\theta+\dot\psi \qquad (4)

where: $\omega_3$ — component along the symmetry axis. This is the single most important formula of top dynamics: the axis component is not the spin rate alone but the spin plus the tilted shadow of the precession. For the symmetric top, $I_3\omega_3$ is the conserved $p_\psi$ ; forgetting the $\dot\phi\cos\theta$ term is the classic route to a wrong precession rate. Collecting (2)–(4) in matrix form exposes the structure — and the pathology — of the map from angle rates to body rates:

\begin{pmatrix}\omega_1\\\omega_2\\\omega_3\end{pmatrix}=\begin{pmatrix}\sin\theta\sin\psi&\cos\psi&0\\\sin\theta\cos\psi&-\sin\psi&0\\\cos\theta&0&1\end{pmatrix}\begin{pmatrix}\dot\phi\\\dot\theta\\\dot\psi\end{pmatrix} \qquad (5)

where: the columns are the body-frame components of $\hat{\mathbf z}$ , $\hat{\mathbf n}$ , $\hat{\mathbf e}_3'$ respectively. Expanding along the third column, $\det = -\sin\theta$ : the matrix is invertible exactly when $\sin\theta \neq 0$ . At $\theta = 0, \pi$ the first and third columns become parallel — $\hat{\mathbf z}$ and $\hat{\mathbf e}_3'$ are the same axis — and the map loses rank: gimbal lock, now as linear algebra. Running the map backwards is what every attitude estimator does with its gyro measurements. Solving (5) for the rates:

\dot\phi=\frac{\omega_1\sin\psi+\omega_2\cos\psi}{\sin\theta},\qquad \dot\theta=\omega_1\cos\psi-\omega_2\sin\psi,\qquad \dot\psi=\omega_3-\dot\phi\cos\theta \qquad (6)

where: the $1/\sin\theta$ in $\dot\phi$ (and, through it, in $\dot\psi$ ) is the price of the chart: near the singular tilts, tiny gyro noise in $\omega_1, \omega_2$ is amplified into wild swings of the reconstructed precession and spin rates, while their sum stays finite. The nutation rate $\dot\theta$ , by contrast, is always well-conditioned — the line of nodes never degenerates. For completeness, the same sum (1) resolved instead onto the space axes gives

\omega_x=\dot\theta\cos\phi+\dot\psi\sin\theta\sin\phi,\quad \omega_y=\dot\theta\sin\phi-\dot\psi\sin\theta\cos\phi,\quad \omega_z=\dot\phi+\dot\psi\cos\theta \qquad (7)

where: the structure mirrors (2)–(4) with the roles of $(\phi, \dot\phi)$ and $(\psi, \dot\psi)$ exchanged — as it must, since viewing the body from space and space from the body are inverse rotations. Use these when tracking $\mathbf L$ or $\boldsymbol\omega$ in the laboratory. One scalar ties the two resolutions together and provides the standard sanity check. The squared magnitude, computed from either set,

\omega^2=\dot\phi^2+\dot\theta^2+\dot\psi^2+2\dot\phi\dot\psi\cos\theta \qquad (8)

where: the sole cross term is $2\dot\phi\dot\psi\,(\hat{\mathbf z}\cdot\hat{\mathbf e}_3') = 2\dot\phi\dot\psi\cos\theta$ — the only pair of the three rotation axes that is not perpendicular. ( $\hat{\mathbf n}$ is perpendicular to both $\hat{\mathbf z}$ and $\hat{\mathbf e}_3'$ by construction, so $\dot\theta$ crosses with nothing.) Any claimed component formulas that fail to reproduce (8) contain an error — this catches most sign mistakes in under a minute.

Go deeper · Derivation B · 2 minBody-frame components by projecting the three rotation axes

Derivation BBody-frame components by projecting the three rotation axes

We are deriving the component formulas (2)–(4) from the master decomposition (1) by resolving each of the three oblique unit vectors — $\hat{\mathbf e}_3'$ , $\hat{\mathbf n}$ , $\hat{\mathbf z}$ — onto the body principal axes. The strategy is to work in decreasing order of ease: the spin axis is a body axis already; the line of nodes needs undoing only the final spin rotation; the space vertical needs the last two rotations undone. The step to watch is the direction of each rotation ( $zxz$ , passive): getting a single sign wrong in the $\psi$ rotation swaps the roles of $\sin\psi$ and $\cos\psi$ and produces formulas that look plausible but fail the $\omega^2$ check (8).

Step 1. The spin axis is trivial:

\hat{\mathbf e}_3'=(0,0,1)_{\text{body}}

Why: $\psi$ is the last rotation and it is performed about the body's own third axis; that axis is a member of the basis we are projecting onto. So $\dot\psi$ feeds only $\omega_3$ .

Step 2. The line of nodes lies in the body equatorial plane. It is the intermediate $x$ -axis — the body $\hat{\mathbf e}_1'$ direction before the final spin by $\psi$ — so in body components it sits at angle $\psi$ behind $\hat{\mathbf e}_1'$ :

\hat{\mathbf n}=(\cos\psi,\,-\sin\psi,\,0)_{\text{body}}

Why the minus sign: the final rotation carried the body axes forward by $\psi$ about $\hat{\mathbf e}_3'$ ; seen from the body, every pre-existing direction appears rotated backward by $\psi$ . So $\dot\theta$ contributes $\dot\theta\cos\psi$ to $\omega_1$ and $-\dot\theta\sin\psi$ to $\omega_2$ .

Step 3. The space vertical needs two rotations undone. Starting from $(0,0,1)$ and applying the tilt and then the spin — equivalently, reading the third column of the full Euler matrix — gives

\hat{\mathbf z}=(\sin\theta\sin\psi,\;\sin\theta\cos\psi,\;\cos\theta)_{\text{body}}

Why the pattern: $\hat{\mathbf z}$ keeps $\cos\theta$ along the symmetry axis (that is the definition of the tilt), and its equatorial shadow of length $\sin\theta$ points at spin phase $\psi$ within the body plane — precisely the geometry of the second figure.

Step 4. Assemble the weighted sum $\boldsymbol\omega = \dot\phi\,\hat{\mathbf z} + \dot\theta\,\hat{\mathbf n} + \dot\psi\,\hat{\mathbf e}_3'$ componentwise:

(\omega_1,\omega_2,\omega_3)=(\dot\phi\sin\theta\sin\psi+\dot\theta\cos\psi,\;\dot\phi\sin\theta\cos\psi-\dot\theta\sin\psi,\;\dot\phi\cos\theta+\dot\psi)

Each column of the matrix in (5) is one of the three resolved unit vectors — the matrix is this derivation in tabular form. Verified: matches Goldstein 3rd ed. eq. (4.87) and Landau–Lifshitz §35; the components reproduce the invariant (8) exactly, and reduce correctly in the pure-spin ( $\boldsymbol\omega = \dot\psi\,\hat{\mathbf e}_3'$ ) and pure-precession ( $\boldsymbol\omega = \dot\phi\,\hat{\mathbf z}$ ) limits.

From kinematics to a Lagrangian

The payoff of the projection is immediate: substitute (2)–(4) into the principal-axis kinetic energy $T = \tfrac12(I_1\omega_1^2 + I_2\omega_2^2 + I_3\omega_3^2)$ and the Euler angles become honest generalized coordinates. For the general (asymmetric) body the expression is bulky, but for the symmetric top ( $I_1 = I_2$ ) — every gyroscope, every toy top, every symmetric molecule — the spin phase cancels out of the transverse terms and the energy collapses to

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

where: $I_1 = I_2$ — transverse principal moments; $I_3$ — axial moment. Neither $\phi$ nor $\psi$ appears undifferentiated: both are cyclic, and that single structural fact is the engine of every top solution that follows in this track.

Go deeper · Derivation C · 1 minSymmetric-top Lagrangian and its two conserved momenta

Derivation CSymmetric-top Lagrangian and its two conserved momenta

We are deriving the conserved momenta $p_\psi$ and $p_\phi$ of the heavy symmetric top from the kinetic energy (9), and assembling them into the effective one-dimensional problem that governs nutation. The strategy is the standard cyclic-coordinate harvest: write $L = T - V$ with the pivot-to-center-of-mass distance $\ell$ giving $V = Mg\ell\cos\theta$ , observe which coordinates are absent, and differentiate. What to watch: $p_\phi$ is not simply $I_1\dot\phi$ — the axial term contributes through the $\dot\phi\cos\theta$ inside $\omega_3$ , and dropping it breaks the conservation law.

Step 1. Write the Lagrangian. With the symmetry axis a distance $\ell$ from the fixed pivot,

L=\tfrac12 I_1(\dot\theta^2+\dot\phi^2\sin^2\theta)+\tfrac12 I_3(\dot\phi\cos\theta+\dot\psi)^2-Mg\ell\cos\theta

Why gravity only brings $\cos\theta$ : the weight acts at the center of mass, whose height above the pivot is $\ell\cos\theta$ ; no other coordinate moves it.

Step 2. Harvest $\psi$ . Since $\partial L/\partial\psi = 0$ ,

p_\psi=\frac{\partial L}{\partial\dot\psi}=I_3(\dot\phi\cos\theta+\dot\psi)=I_3\,\omega_3

is conserved. Why this is physical: no torque component along the symmetry axis exists (gravity's torque is along the line of nodes), so the axial angular momentum cannot change.

Step 3. Harvest $\phi$ . Since $\partial L/\partial\phi = 0$ ,

p_\phi=\frac{\partial L}{\partial\dot\phi}=I_1\dot\phi\sin^2\theta+I_3(\dot\phi\cos\theta+\dot\psi)\cos\theta = I_1\dot\phi\sin^2\theta + p_\psi\cos\theta

is conserved. Why this is $L_z$ : gravity pulls straight down, so its torque has no vertical component and the lab-vertical angular momentum survives. Note the second term — the axial spin leaks into $L_z$ through the tilt.

Step 4. Reduce to one dimension. Solve Step 3 for $\dot\phi = (p_\phi - p_\psi\cos\theta)/(I_1\sin^2\theta)$ and substitute into the conserved energy:

E=\tfrac12 I_1\dot\theta^2+\underbrace{\frac{(p_\phi-p_\psi\cos\theta)^2}{2I_1\sin^2\theta}+\frac{p_\psi^2}{2I_3}+Mg\ell\cos\theta}_{V_{\mathrm{eff}}(\theta)}

The tilt $\theta$ now moves in a one-dimensional effective potential whose wells and turning points are the steady precession and the nutation bounds analyzed in the heavy-top article — this derivation is the doorway to it. Verified: reproduces the Goldstein §5.7 reduction; setting $\dot\theta = 0$ , $\ddot\theta = 0$ in the $\theta$ equation recovers the standard steady-precession condition, and for the fast top yields $\dot\phi \approx Mg\ell/(I_3\omega_3)$ .

Which formula answers which question

you want	use	notes
the exact, convention-free statement	$\boldsymbol\omega=\dot\phi\hat{\mathbf z}+\dot\theta\hat{\mathbf n}+\dot\psi\hat{\mathbf e}_3'$ , eq. (1)	valid always; components need a convention
body components from angle rates	eqs. (2)–(4) or matrix (5)	$zxz$ convention; feeds $T$ , $\mathbf L$ , Euler's equations
angle rates from gyro measurements	inverse (6)	blows up as $1/\sin\theta$ near lock
lab-frame components	eq. (7)	$\phi \leftrightarrow \psi$ mirror of the body set
a fast error check	invariant $\omega^2$ , eq. (8)	one cross term, $2\dot\phi\dot\psi\cos\theta$
top dynamics setup	$T$ from (9), then the two momenta	$\phi, \psi$ cyclic — the whole game

Limiting cases

Case	Result	Meaning
pure spin, $\dot\phi=\dot\theta=0$	$(\omega_1,\omega_2,\omega_3)=(0,0,\dot\psi)$	the body turns only about its symmetry axis
steady precession, $\dot\theta=0$	$\omega_3=\dot\phi\cos\theta+\dot\psi$ , $\omega_\perp=\dot\phi\sin\theta$	transverse component of constant magnitude circulates in the body at rate $-\dot\psi$
gimbal lock, $\theta\to0$	$\omega_3\to\dot\phi+\dot\psi$ ; $\det\to0$	only the sum of precession and spin is physical; inverse map fails as $1/\sin\theta$
spherical top, $I_1=I_2=I_3$	$T=\tfrac12 I\omega^2$ from (8)	all $\psi$ -dependence gone; any axes are principal
slow rates, $\dot\phi\dot\psi \ll \dot\theta^2$	$\omega^2 \approx \dot\phi^2+\dot\theta^2+\dot\psi^2$	the oblique axes act effectively orthogonal when the only non-perpendicular pair barely co-rotates

Common trapThe three Euler rates are just the components of omega in disguise

They are not, and treating $(\dot\phi, \dot\theta, \dot\psi)$ as orthogonal Cartesian components of $\boldsymbol\omega$ is the most damaging habit a student of rigid bodies can carry. The three rates turn about three axes that are mutually oblique ( $\hat{\mathbf z}\cdot\hat{\mathbf e}_3' = \cos\theta \neq 0$ ), belong to different frames, and move relative to both. Concretely: $\omega^2 \neq \dot\phi^2 + \dot\theta^2 + \dot\psi^2$ (the cross term $2\dot\phi\dot\psi\cos\theta$ is missing), and $\omega_3 \neq \dot\psi$ (the precession leaks $\dot\phi\cos\theta$ into the axis component). A second trap hides one level deeper: there exists no set of three orientation coordinates whose time derivatives are the components of $\boldsymbol\omega$ — the angular velocity is a non-integrable (anholonomic) combination, which is exactly why the projection matrix in (5) depends on the angles themselves. Always project via (2)–(4); never shortcut.

Worked examples

1 · A toy gyroscope in steady precession (confidence builder)

A toy gyroscope precesses steadily at $\dot\phi = 2.0\ \mathrm{rad/s}$ at a fixed tilt $\theta = 30°$ while spinning at $\dot\psi = 50\ \mathrm{rad/s}$ . Its moments are $I_1 = I_2 = 4.0\times10^{-5}\ \mathrm{kg\,m^2}$ , $I_3 = 6.0\times10^{-5}\ \mathrm{kg\,m^2}$ . Find the body-frame components of $\boldsymbol\omega$ at the instant $\psi = 0$ , the axial angular momentum, and the kinetic energy.

Step 1. Components from (2)–(4) with $\dot\theta = 0$ , $\sin 30° = 0.5$ , $\cos 30° = 0.8660$ , $\psi = 0$ : $\omega_1 = \dot\phi\sin\theta\sin\psi = 0$ ; $\omega_2 = \dot\phi\sin\theta\cos\psi = 2.0\times0.5 = \boxed{1.00\ \mathrm{rad/s}}$ ; $\omega_3 = \dot\phi\cos\theta + \dot\psi = 1.732 + 50 = \boxed{51.73\ \mathrm{rad/s}}$ .

Step 2. Check with the invariant (8): $\omega^2 = \dot\phi^2 + \dot\psi^2 + 2\dot\phi\dot\psi\cos\theta = 4 + 2500 + 173.2 = 2677.2\ \mathrm{rad^2/s^2}$ , so $\omega = 51.74\ \mathrm{rad/s}$ . Direct sum: $0^2 + 1.00^2 + 51.73^2 = 2677.2$ ✓ — the two routes agree to four figures.

Step 3. Axial angular momentum (the conserved $p_\psi$ ): $L_3 = I_3\omega_3 = 6.0\times10^{-5}\times51.73 = \boxed{3.10\times10^{-3}\ \mathrm{kg\,m^2/s}}$ . Note it is $3.5\%$ larger than the naive $I_3\dot\psi = 3.00\times10^{-3}$ : the precession's shadow $\dot\phi\cos\theta$ is small here but not negligible, and in slower-spinning tops it dominates.

Step 4. Kinetic energy from (9): $T = \tfrac12 I_1\dot\phi^2\sin^2\theta + \tfrac12 I_3\omega_3^2 = \tfrac12(4.0\times10^{-5})(1.00) + \tfrac12(6.0\times10^{-5})(2676.2) = 2.0\times10^{-5} + 8.029\times10^{-2} = \boxed{8.03\times10^{-2}\ \mathrm J}$ . The spin term carries $99.97\%$ of the energy — the "fast top" regime in numbers.

2 · Reconstructing angle rates from gyro data, and the price near lock (exam level)

A strapdown IMU on a spinning projectile reports body rates $(\omega_1, \omega_2, \omega_3) = (0.10,\ 0.20,\ 1.50)\ \mathrm{rad/s}$ when the attitude solution reads $\theta = 60°$ , $\psi = 30°$ . Recover $(\dot\phi, \dot\theta, \dot\psi)$ , verify, and repeat the $\dot\phi$ reconstruction at $\theta = 2°$ with the same body rates to exhibit the singularity.

Step 1. Precession rate from (6), with $\sin\psi = 0.5$ , $\cos\psi = 0.8660$ , $\sin 60° = 0.8660$ :

\dot\phi = \frac{0.10\times0.5 + 0.20\times0.8660}{0.8660} = \frac{0.2232}{0.8660} = \boxed{0.2578\ \mathrm{rad/s}}

Step 2. Nutation rate: $\dot\theta = \omega_1\cos\psi - \omega_2\sin\psi = 0.0866 - 0.100 = \boxed{-0.0134\ \mathrm{rad/s}}$ — the tilt is closing slowly.

Step 3. Spin rate: $\dot\psi = \omega_3 - \dot\phi\cos\theta = 1.50 - 0.2578\times0.5 = \boxed{1.371\ \mathrm{rad/s}}$ .

Step 4. Verify by running (2)–(4) forward: $\omega_1 = 0.2578\times0.8660\times0.5 + (-0.0134)\times0.8660 = 0.1116 - 0.0116 = 0.1000$ ✓; $\omega_2 = 0.2578\times0.8660\times0.8660 - (-0.0134)\times0.5 = 0.1933 + 0.0067 = 0.2000$ ✓; $\omega_3 = 0.1289 + 1.371 = 1.500$ ✓.

Step 5. Now the same body rates at $\theta = 2°$ ( $\sin\theta = 0.0349$ ): $\dot\phi = 0.2232/0.0349 = \boxed{6.40\ \mathrm{rad/s}}$ and $\dot\psi = 1.50 - 6.40\times0.9994 = -4.89\ \mathrm{rad/s}$ — the reconstructed precession has grown $25$ -fold and the spin has flipped sign, while their sum $\dot\phi + \dot\psi = 1.50\ \mathrm{rad/s}$ (the only physical combination near lock) stays tame. The amplification factor is exactly $1/\sin 2° = 28.7$ : any gyro noise in $\omega_1, \omega_2$ is magnified by the same factor. This is why flight software integrates quaternions and converts to Euler angles only at the display layer.

Origins · 1 minArnold Sommerfeld and how this was discovered

Arnold Sommerfeld

1868–1951 · Munich

Before Sommerfeld became the great teacher of quantum theory, he spent a decade inside the spinning top. With Felix Klein he wrote the four-volume Über die Theorie des Kreisels (1897–1910) — nearly two thousand pages that remain the most complete anatomy of rotating rigid bodies ever produced, working out the kinematics of Euler angles, the top's integrable cases, and, in the final volume, the technical applications: gyrocompasses, ship stabilizers, the ballistics of spinning shells. That engineering seriousness became his trademark. At Munich he built the most productive school in the history of theoretical physics — Heisenberg, Pauli, Debye, and Bethe were his doctoral students, and he was nominated for the Nobel Prize a record 84 times without ever winning it, though four of his students did. His fine-structure constant $\alpha$ and the Bohr–Sommerfeld quantization rules carried the top's action–angle machinery straight into the quantum theory of the atom.

History

1760

— Euler derives the kinematic equations relating the rates of his three angles to the angular velocity of a rigid body.

1788

— Lagrange's Mécanique analytique treats the heavy symmetric top with

(\phi,\theta,\psi)

as generalized coordinates, exhibiting the two cyclic momenta.

1852

— Foucault builds and names the gyroscope, turning Euler-angle kinematics into a laboratory instrument for seeing the Earth rotate.

1897

— Klein and Sommerfeld begin the four-volume Theory of the Top, the definitive treatment of rigid-body kinematics and its applications.

1969

— Apollo's gimballed inertial platform mechanizes the Euler-angle chart in hardware, and its gimbal-lock zone becomes an operational flight constraint.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics (3rd ed.), §4.4 and §4.9 · Landau & Lifshitz, Mechanics (3rd ed.), §35 · Morin, Introduction to Classical Mechanics, Ch. 9 · Taylor, Classical Mechanics, Ch. 10

Provenance. Equations to recompute by a verifier: the body-frame components (Goldstein 3rd ed. eq. 4.87; Landau–Lifshitz Mechanics §35) for the z–x–z convention; the determinant $-\sin\theta$ of the rate matrix; $\omega^2=\dot\phi^2+\dot\theta^2+\dot\psi^2+2\dot\phi\dot\psi\cos\theta$ ; the symmetric-top kinetic energy and conserved momenta $p_\psi,p_\phi$ (Goldstein §5.7). Conventions: z–x–z (Goldstein/Landau). No empirical numbers used; molecular-rotor energy expression $E_{JK}=BJ(J+1)+(A-B)K^2$ is the standard symmetric-top formula (Herzberg). Worked-example numbers on this page ( $\omega_3 = 51.73\ \mathrm{rad/s}$ , $T = 8.03\times10^{-2}\ \mathrm J$ ; the reconstruction $(\dot\phi,\dot\theta,\dot\psi) = (0.2578, -0.0134, 1.371)\ \mathrm{rad/s}$ and the $1/\sin 2° = 28.7$ amplification) are illustrative arithmetic recomputed from equations (2)–(9).

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

If finite rotations do not commute, why exactly are we allowed to add the three omega arrows?

Because addition happens at first order in time. Write each infinitesimal rotation as $\mathbb 1 + \mathrm d\alpha\,[\hat{\mathbf a}]_\times$ ; composing three of them multiplies the brackets, and every order-dependent term carries a product of two small angles — $\mathrm d\phi\,\mathrm d\theta$ and the like. Divide by $\mathrm dt$ and let $\mathrm dt \to 0$ : those crossed terms scale as $\mathrm dt$ and vanish, leaving the generators to add cleanly. Non-commutativity is perfectly real — it is why your final orientation depends on the path of rotations, and why $\omega$ cannot be the derivative of any orientation coordinates — but it lives at second order, invisible to an instantaneous rate. A concrete check: two successive $90°$ turns about different axes disagree with the reversed order by a $120°$ rotation, while two $1°$ turns disagree by only $0.0175°$ — quadratically small.

Is the angular velocity the time derivative of some rotation vector, the way linear velocity is the derivative of position?

No — and this is one of the deepest structural facts of rigid-body mechanics. There exists no triple of orientation coordinates whose time derivatives equal $(\omega_1, \omega_2, \omega_3)$ ; the relation (5) between $\boldsymbol\omega$ and the Euler rates involves the angles themselves and cannot be integrated away. Technically, $\omega_i\,\mathrm dt$ are non-exact (anholonomic) differentials. You can see it physically: if $\boldsymbol\omega$ were $\mathrm d\mathbf s/\mathrm dt$ for some coordinates $\mathbf s$ , then a closed loop of rotations returning to the start would require $\oint\boldsymbol\omega\,\mathrm dt = 0$ — but you can easily rotate a book through a sequence with nonzero net $\oint\boldsymbol\omega\,\mathrm dt$ that returns it exactly to its initial orientation. This anholonomy is why orientation must be integrated as a matrix or quaternion, not as three accumulating angles.

Why project onto the body frame rather than the space frame — and when would I want the space components?

Project onto whichever basis makes your dynamics diagonal. The inertia tensor is constant and diagonal in the body principal axes, so $T = \tfrac12\sum I_i\omega_i^2$ and Euler's equations take their clean forms only there — that is why (2)–(4) are the default. The space components (7) earn their keep when the conserved quantity lives in the lab: tracking the vertical component $L_z$ of angular momentum, describing what a stationary camera sees, or matching boundary conditions fixed in the room. The two sets are related by the full Euler rotation matrix, and computing $\omega^2$ from each — both must give (8) — is the standard cross-check that your conventions are consistent.

Why is the axis component of omega not just the spin rate psi-dot?

Because the body axis is tilted, and the precession leaks into it. The rotation $\dot\phi$ happens about the vertical $\hat{\mathbf z}$ , and $\hat{\mathbf z}$ has projection $\cos\theta$ onto the symmetry axis — so the body is being carried around its own axis at rate $\dot\phi\cos\theta$ in addition to its spin $\dot\psi$ , giving $\omega_3 = \dot\phi\cos\theta + \dot\psi$ . The distinction has teeth: the conserved momentum of a symmetric top is $p_\psi = I_3\omega_3$ , not $I_3\dot\psi$ , and the steady-precession formula $\dot\phi \approx Mg\ell/(I_3\omega_3)$ uses the full $\omega_3$ . In the worked example above the difference was $3.5\%$ ; for a slowly spinning, fast-precessing top it can dominate. Only at $\theta = 90°$ — axis horizontal — does $\omega_3 = \dot\psi$ hold exactly.

What physically goes wrong in the rate equations at gimbal lock, and why does the nutation rate stay well-behaved?

At $\theta = 0$ the vertical and the body axis coincide, so precession and spin are rotations about the same line: the body's motion determines their sum $\dot\phi + \dot\psi$ but contains no information about the split. The matrix (5) expresses this by dropping to rank 2 ( $\det = -\sin\theta \to 0$ ), and the inverse (6) manufactures the missing information out of noise — hence the $1/\sin\theta$ amplification of $\dot\phi$ and $\dot\psi$ individually, with their sum remaining finite and accurate. The nutation rate is immune because its axis, the line of nodes, is perpendicular to both $\hat{\mathbf z}$ and $\hat{\mathbf e}_3'$ and never degenerates with either: $\dot\theta = \omega_1\cos\psi - \omega_2\sin\psi$ contains no $1/\sin\theta$ . The lock is a property of the chart, not the body — the physical motion through $\theta = 0$ is perfectly smooth.

Why do the psi cross-terms cancel when I add the squares of the first two components?

Because $\omega_1$ and $\omega_2$ are the components of one fixed transverse vector read off in a basis that merely spins with phase $\psi$ . Expanding, $\omega_1^2 + \omega_2^2 = (\dot\phi\sin\theta)^2(\sin^2\psi + \cos^2\psi) + \dot\theta^2(\cos^2\psi + \sin^2\psi) + 2\dot\phi\dot\theta\sin\theta(\sin\psi\cos\psi - \cos\psi\sin\psi)$ — the cross terms are equal and opposite by construction, leaving $\dot\phi^2\sin^2\theta + \dot\theta^2$ . Geometrically: the transverse part of $\boldsymbol\omega$ has two orthogonal contributions, $\dot\phi\sin\theta$ (shadow of precession) and $\dot\theta$ (along the nodes), and rotating the basis by $\psi$ cannot change the length of their resultant. This is also why the symmetric-top energy (9) is $\psi$ -free — the deep reason $\psi$ is cyclic.

How do real navigation systems integrate orientation if the Euler-rate equations are singular?

They sidestep the chart entirely. Flight software carries the attitude as a unit quaternion $q$ and integrates $\dot q = \tfrac12\,q \otimes (0, \boldsymbol\omega)$ , where $\boldsymbol\omega$ is the gyro triad's body-rate measurement — a bilinear equation with constant coefficients, no trigonometric functions, and no singular configuration anywhere. The unit-norm constraint is renormalized each step to kill drift. Euler angles are computed from the quaternion only when a human needs them (cockpit displays, telemetry), accepting the $1/\sin\theta$ conditioning at the display layer where it can do no dynamical harm. The lesson generalizes: equations (2)–(6) are the right tool for understanding and for analytical mechanics; the quaternion is the right tool for numerical propagation. Professionals keep both and convert at the interfaces.