Inertia Ellipsoid

Spacecraft engineers live on this ellipsoid. Explorer 1's 1958 flat-spin surprise — a slender body spun about its minimum-moment axis, flexing antennas bleeding energy until the spin migrated to the maximum-moment axis — is now a textbook design rule ("major-axis rule": a dissipating free spinner is stable only about the largest principal moment), and every satellite's mass-properties report ships the inertia tensor whose ellipsoid attitude-control teams visualize before choosing spin axes. The polhode is not a museum piece either: Gravity Probe B's ultra-spherical gyroscopes (2004–2005) showed slow polhode wander that modulated their readout and had to be modeled to milliarcsecond precision, and mission teams tracking tumbling debris or a dead satellite for capture first fit its polhode state — which curve on which inertia ellipsoid — from light-curve flicker before daring an approach. The deeper legacy is the representation quadric habit: picture any symmetric rank-2 tensor as its ellipsoid. Crystal optics reads refractive indices off the index ellipsoid (the optical indicatrix); continuum mechanics draws stress and strain ellipsoids descended from Cauchy's original; diffusion-MRI tractography renders millions of brain-voxel diffusion tensors as tiny glyph ellipsoids whose long axes trace nerve-fiber bundles. Every one of those pictures works for exactly the reason the inertia ellipsoid works: real + symmetric + positive-definite means a bounded ellipsoid whose axes are eigenvectors — a theorem you can now watch roll across a table.

Stand at the body's reference point and look out along some direction $\hat{\mathbf n}$. The body is hard to spin about that axis if mass lies far from it — large $I(\hat{\mathbf n})$ — and easy to spin if the mass hugs it. Now do this for every direction at once, and record your findings as a surface: walk out a distance $1/\sqrt{I}$ along each axis and leave a marker. The markers close into a smooth, egg-shaped skin around you. That skin is the inertia ellipsoid: a mass-distribution portrait turned inside out, bulging OUT where the body is easy to spin and tucking IN where it is hard.

Three facts give the picture its power. First, the skin is rigid: the tensor's components are constants in the body frame, so the ellipsoid is frozen into the body and tumbles with it, like a ghostly exoskeleton. Second, it is complete: the moment of inertia about every conceivable axis through the point — infinitely many numbers — is stored as the surface's radius function, and conversely six well-chosen radius measurements reconstruct all six independent components of $\mathbf I$. Third, it is honest about orientation: if your coordinate axes are not principal axes, the ellipsoid sits visibly askew in them, and the off-diagonal products of inertia are precisely the algebraic record of that skew. The shape also tells you, before any dynamics, what kind of rotator you are holding. All three semi-axes equal: a spherical top, every axis principal, no wobble possible. Two equal: a symmetric top — coin, pencil, frisbee — with one distinguished axis and a plane of interchangeable ones. All three different: an asymmetric top, the home of the intermediate-axis instability that flips tennis rackets and phones. Diagonalizing a $3\times 3$ symmetric matrix can feel like dry linear algebra; here it is literally finding the symmetry axes of an oval.

One warning before the equations, because it catches everyone once: the radius is $1/\sqrt{I}$, an inverse. Long ellipsoid axes are EASY spin axes. A thin rod's ellipsoid is a long cigar along the rod (tiny moment about its own axis, huge radius); a coin's is a squashed bun, flattest along the coin's symmetry axis (largest moment, shortest radius). The ellipsoid resembles the body's proportions in orientation, never in scale — and reading it the wrong way round inverts every conclusion you draw from it.

Everything starts from the scalar moment of inertia about an arbitrary axis, delivered by the tensor article as the quadratic form evaluated on the direction. If you spin the body at $\boldsymbol\omega = \omega\hat{\mathbf n}$, the kinetic energy is $\tfrac12 I(\hat{\mathbf n})\,\omega^2$ with

where: $I(\hat{\mathbf n})$ — scalar moment of inertia about the axis through the origin along $\hat{\mathbf n}$; $\hat{\mathbf n}$ — unit vector specifying the rotation axis, $n_in_i=1$; $I_{ij}$ — components of the inertia tensor about the chosen origin. Read it as a price list: one number per direction, quoting what a unit of spin costs in angular momentum and energy. In the principal frame it expands to $I_1n_1^2 + I_2n_2^2 + I_3n_3^2$ — a weighted average of the principal moments, which is why $I(\hat{\mathbf n})$ can never escape the bracket $[I_1, I_3]$. The ellipsoid packages this entire price list as one surface. Take the equation $I(\mathbf x) \cdot |\mathbf x|^2$-style and freeze its value at unity:

where: $x_i$ — Cartesian coordinates of a point on the surface, measured from the reference point (their unit is $1/\sqrt{\text{kg}\,\text{m}^2}$ — this is a surface in an abstract inertia space, not a physical shell in meters); $I_{ij}$ — the inertia tensor; its positive-definiteness is what guarantees a bounded ellipsoid, as Derivation A proves. A point of the surface in direction $\hat{\mathbf n}$ sits at distance $r = 1/\sqrt{I(\hat{\mathbf n})}$ — substitute $x_i = r\,n_i$ into (2) and the rule falls out in one line. Rotate your axes onto the ellipsoid's own symmetry axes and the cross terms die. What remains is the standard form every analytic-geometry course would recognize:

where: $I_i$ — the principal moments (eigenvalues of $\mathbf I$); $a_i$ — the semi-axis lengths; $x_i$ — coordinates along the principal (body) axes. This is the dictionary entry of the whole subject: eigenvalues ↔ semi-axes, eigenvectors ↔ symmetry axes. A uniform thin disk, with $I_3 = 2I_1$ by the perpendicular axis theorem, has an ellipsoid exactly $\sqrt2$ times wider than it is tall — a fixed shape, whatever the disk's mass or size. The algebraic engine behind the dictionary is the spectral decomposition of the symmetric tensor,

where: $R$ — the orthogonal matrix rotating lab axes into principal axes; $\mathrm{diag}(I_1,I_2,I_3)$ — the tensor in its own frame. Geometrically, $R$ is nothing but the instruction "turn your head until the ellipsoid looks upright." When the lab axes are not principal, the off-diagonal entries of $\mathbf I$ are nonzero and the ellipsoid sits tilted in your frame — Derivation C quantifies the tilt angle and connects it to the wheel-balancing problem.

So far the ellipsoid is a portrait. Poinsot's discovery is that the portrait moves — that free rotation is a rolling motion of this very surface. The bridge from statics to dynamics is a pair of innocent-looking equations. First, the kinetic energy of an arbitrary spin, inherited from the tensor article:

where: $T$ — rotational kinetic energy; $\boldsymbol\omega$ — angular velocity. Read the second form geometrically: at fixed energy, the allowed angular velocities sweep out a surface in $\boldsymbol\omega$-space — the energy ellipsoid — which is the inertia ellipsoid scaled by the factor $\sqrt{2T}$. Same shape, same axes, different size and different space: its coordinates are angular velocities, not abstract lengths. Second, the fact that unlocks everything. Differentiate the energy with respect to the angular velocity, using the symmetry $I_{ij}=I_{ji}$, and the angular momentum appears as a gradient:

where: $\mathbf L$ — angular momentum about the fixed point; $\nabla_{\!\boldsymbol\omega}$ — gradient in angular-velocity space. A gradient is always perpendicular to the level surfaces of its function — so $\mathbf L$ is normal to the energy ellipsoid at the tip of $\boldsymbol\omega$. This is the cleanest way to see why $\mathbf L$ and $\boldsymbol\omega$ are generally not parallel: $\boldsymbol\omega$ is a radius of the ellipsoid, $\mathbf L$ is the surface normal where that radius lands, and on a non-spherical surface a radius and its normal align only on the symmetry axes — the principal axes, exactly as the angular-momentum article computes algebraically. Two scalar companions complete the toolkit. When the spin is locked along a principal direction $\hat{\mathbf n}$, energy and momentum trade through the radius:

where: $L$ — magnitude of $\mathbf L$ for $\boldsymbol\omega \parallel \hat{\mathbf n}$ a principal direction; $T$ — kinetic energy. At fixed energy, spinning about the large-moment axis stores the most angular momentum — the fact that decides where a dissipating spinner ends up, as the Explorer 1 FAQ below spells out. And in torque-free motion, where both $T$ and $\mathbf L$ are conserved, the projection of $\boldsymbol\omega$ onto the fixed direction $\hat{\mathbf L}$ is pinned:

where: $d$ — the constant height of the tip of $\boldsymbol\omega$ above the plane through the origin perpendicular to $\mathbf L$; $2T, L$ — the two conserved quantities of free rotation. This constant is the invariable plane: a plane fixed in space, at distance $d/\sqrt{2T} = \sqrt{2T}/L$ from the center in the scaled picture, on which the whole motion will now be made to roll.

Which equation answers which question

When the products of inertia vanish, the ellipsoid stands upright in your axes and life is simple. When they do not, the ellipsoid leans — and the lean angle is exactly the rotation that diagonalizes the tensor. The products-of-inertia article does the full three-dimensional bookkeeping; here is the two-dimensional core, which is also the mechanic's wheel-balancing formula.

The ellipsoid's shape mirrors the spectrum of $\mathbf I$:

1 · The ellipsoid of a spinning disk (confidence builder)

A uniform disk has mass $M = 0.50\ \mathrm{kg}$ and radius $R = 0.12\ \mathrm{m}$. Build its inertia ellipsoid about the center of mass, and find the moment of inertia about an axis through the center at $45°$ to the symmetry axis.

Step 1. Principal moments. For a uniform disk the symmetry axis and any two diameters are principal, with $I_3 = \tfrac12 MR^2 = \tfrac12 (0.50)(0.12)^2 = 3.6\times10^{-3}\ \mathrm{kg\,m^2}$ and, by the perpendicular axis theorem, $I_1 = I_2 = \tfrac14 MR^2 = 1.8\times10^{-3}\ \mathrm{kg\,m^2}$.

Step 2. Semi-axes of the ellipsoid. Using $a_i = 1/\sqrt{I_i}$: $a_1 = a_2 = 1/\sqrt{1.8\times10^{-3}} = 23.6$ and $a_3 = 1/\sqrt{3.6\times10^{-3}} = 16.7$ (units $\mathrm{(kg\,m^2)^{-1/2}}$). The ellipsoid is an oblate spheroid, squashed along the disk's symmetry axis, with the universal disk ratio $a_1/a_3 = \sqrt{2} \approx \boxed{1.414}$ — independent of $M$ and $R$.

Step 3. Read off a tilted-axis moment. For $\hat{\mathbf n}$ at $45°$ to the symmetry axis, $n_3^2 = n_\perp^2 = \tfrac12$, so equation (1) gives

Step 4. Audit with the radius rule. The ellipsoid radius along that direction should be $1/\sqrt{2.7\times10^{-3}} = 19.2$ — and indeed $16.7 < 19.2 < 23.6$: the tilted radius sits between the two semi-axes, exactly as a point between the pole and the equator of a spheroid must. The price list interpolates; it never extrapolates.

2 · A tumbling brick: where L really points (exam level)

A rectangular book of mass $M = 1.2\ \mathrm{kg}$ and dimensions $0.20 \times 0.15 \times 0.03\ \mathrm{m}$ spins at $\omega = 30\ \mathrm{rad/s}$ about its face diagonal (in the plane of the covers). Find the moment of inertia about that axis, the angle between $\mathbf L$ and $\boldsymbol\omega$, and the Poinsot constant $2T/L$.

Step 1. Principal moments of a uniform block, $I = \tfrac{M}{12}(\text{sum of squares of the two transverse dimensions})$: $I_1 = \tfrac{1.2}{12}(0.15^2+0.03^2) = 2.34\times10^{-3}$, $I_2 = \tfrac{1.2}{12}(0.20^2+0.03^2) = 4.09\times10^{-3}$, $I_3 = \tfrac{1.2}{12}(0.20^2+0.15^2) = 6.25\times10^{-3}\ \mathrm{kg\,m^2}$. Triaxial: an asymmetric top, with ellipsoid semi-axes $20.7$, $15.6$, $12.6$.

Step 2. The face diagonal of the $0.20\times0.15$ face has direction $\hat{\mathbf n} = (0.20, 0.15, 0)/0.25 = (0.8, 0.6, 0)$. The quadratic form (1):

Step 3. Angular momentum. With $\boldsymbol\omega = 30\,(0.8, 0.6, 0)\ \mathrm{rad/s}$, each principal component gets its own price: $\mathbf L = (I_1\omega_1,\ I_2\omega_2,\ 0) = (0.0562,\ 0.0736,\ 0)\ \mathrm{kg\,m^2/s}$, so $L = 0.0926\ \mathrm{kg\,m^2/s}$. The angle to $\boldsymbol\omega$ comes from $\cos\alpha = (\mathbf L\cdot\hat{\mathbf n})/L = (2.97\times10^{-3}\times30)/0.0926 = 0.962$, giving $\alpha = \boxed{15.8°}$. Geometrically: $\boldsymbol\omega$ pierces the ellipsoid off any symmetry axis, and $\mathbf L$ tips away from it toward the hard ($I_2$) direction — the surface normal leans toward the shorter semi-axis.

Step 4. Poinsot constant. $T = \tfrac12 I(\hat{\mathbf n})\,\omega^2 = \tfrac12 (2.97\times10^{-3})(900) = 1.34\ \mathrm{J}$, so $2T/L = 2.67/0.0926 = \boxed{28.9\ \mathrm{rad/s}}$.

Step 5. Audit. Equation (8) promises $2T/L = \hat{\mathbf L}\cdot\boldsymbol\omega = \omega\cos\alpha = 30 \times 0.962 = 28.9\ \mathrm{rad/s}$ ✓. If this book were released torque-free with exactly this spin, the tip of $\boldsymbol\omega$ would stay $28.9\ \mathrm{rad/s}$ above the invariable plane forever while wandering along its polhode — and since the face diagonal is no principal axis, the book would visibly wobble: $\mathbf L$ stays fixed in space, so $\boldsymbol\omega$ (15.8° away from it) must precess around it.

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §5.4 (the inertia ellipsoid) and §5.6 (Poinsot construction) · Landau & Lifshitz, Mechanics, Vol. 1, §32 (the inertia ellipsoid and free rotation) · Morin, Introduction to Classical Mechanics, Ch. 9 (angular momentum, principal axes) · Taylor, Classical Mechanics, Ch. 10 (rotational motion of rigid bodies) All equations are standard. The key checkpoints are: (i) $r=1/\sqrt{I(\hat{\mathbf n})}$ follows by substituting $x_i=rn_i$ into $I_{ij}x_ix_j=1$; (ii) semi-axes $a_i=1/\sqrt{I_i}$ come from the principal-axis form of the quadric; (iii) $\nabla_{\!\boldsymbol\omega}(\tfrac12\boldsymbol\omega^\top\mathbf I\boldsymbol\omega)=\mathbf I\boldsymbol\omega=\mathbf L$ uses the symmetry of $\mathbf I$; (iv) positive-definiteness yields a bounded ellipsoid by the spectral theorem. The only numerical constant introduced independently is the disk relation $I_3=2I_1$, which follows from the perpendicular axis theorem for a uniform thin disk. Worked-example numbers on this page (disk: $I_3 = 3.6\times10^{-3}$, $I_1 = 1.8\times10^{-3}\ \mathrm{kg\,m^2}$, $a_1/a_3=\sqrt2$; brick: $I(\hat{\mathbf n}) = 2.97\times10^{-3}\ \mathrm{kg\,m^2}$, misalignment $15.8°$, $2T/L = 28.9\ \mathrm{rad/s}$) are illustrative arithmetic recomputed from the boxed formulas.

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