Fermi-Walker Transport

Parallel transport answers a sharp question: how do I carry a vector along a curve while changing it as little as the connection allows? Along a geodesic that is the whole story — free-fall is the state of no forces, and a gyroscope in free fall keeps its axis parallel-transported. But most observers are not in free fall. A rocket burns fuel, a charge feels the electromagnetic field, a bead is held on a spinning wire, a satellite's clock rides a bent orbit. Their worldlines are accelerated, and on an accelerated worldline naive parallel transport breaks a promise it never advertised: it fails to carry the observer's own four-velocity into the new four-velocity. The tangent slides off the motion. Fermi-Walker transport is the minimal repair. It keeps everything parallel transport got right — still an isometry, still built only from the metric connection, still reducing to parallel transport the instant the acceleration vanishes — but adds exactly one thing: the infinitesimal Lorentz boost, in the plane spanned by velocity and acceleration, needed to keep the four-velocity transported into itself. In particular it adds no spatial rotation, which is why it describes an ideal torque-free gyroscope. The residual drift a carried gyroscope shows against the distant stars — Thomas precession in flat spacetime, geodetic (de Sitter) precession in a gravitational field — is not extra input; it is the failure of successive boosts to compose back to the identity. This one law, drawn from aq-gr-parallel-transport and built on aq-gr-metric-compatibility, is what lets an accelerated observer carry rulers, clocks, and spin axes, and it is the scaffold of Fermi normal coordinates.

Key takeaways

Fermi-Walker transport carries a vector along any timelike worldline — accelerated or not — so it stays as rotation-free as the motion permits, keeping the four-velocity transported into itself.
On a geodesic the four-acceleration is zero and the law collapses exactly to parallel transport; the two agree wherever there is no proper acceleration.
The extra term is antisymmetric, so the transport is an isometry: inner products, norms, and the orthonormality of a carried tetrad are all conserved along the curve.
Physically it is the law of a torque-free gyroscope: the spin axis is Fermi-Walker transported, so its drift against a locally inertial frame is the geometric Thomas or geodetic precession.
The generator is the boost Ω = u∧a in the velocity–acceleration plane — no spatial rotation — which is why an added spatial spin turns a Fermi-Walker frame into a deliberately rotating one.
Key idea: swap the parallel-transport condition DV/dτ = 0 for DV^μ/dτ = (u^μ a_ν − a^μ u_ν)V^ν. The added antisymmetric boost is the unique rotation that keeps the four-velocity carried into itself while preserving every inner product, and it switches off exactly when the acceleration does.

The core insight

Replace the parallel-transport condition DV^μ/dτ = 0 by DV^μ/dτ = (u^μ a_ν − a^μ u_ν)V^ν; the added antisymmetric term is the unique infinitesimal Lorentz boost that keeps the four-velocity transported into itself while preserving all inner products, and it vanishes precisely when the four-acceleration does, recovering parallel transport on geodesics.

Where this lives today

The most literal test flew from 2004 to 2005: Gravity Probe B, a Stanford–NASA satellite carrying four fused-quartz gyroscopes machined to be among the roundest objects ever made. Each spin axis is (on the geodesic orbit) parallel-transported, and the mission measured the geodetic precession of $6601.8\pm18.3$ mas/yr and, harder still, the tiny frame-dragging precession of $37.2\pm7.2$ mas/yr from Earth's rotation — a second, independent piece of the same transport story. Beyond that landmark the mathematics is working machinery. Fermi normal coordinates built on a Fermi-Walker triad are the standard local frame for modeling how a laser-interferometer gravitational-wave detector responds to a passing wave, and for defining what "proper distance between the mirrors" even means in a curved, time-varying spacetime. The spin equation generalizes, with curvature coupling, to the Mathisson–Papapetrou–Dixon equations governing how a spinning compact body deviates from geodesic motion — a live concern for the extreme-mass-ratio inspirals future space detectors aim to catch. The same Thomas precession that fixes atomic fine structure reappears in storage-ring spin dynamics, where the muon $g\!-\!2$ experiments must account for it to part-per-million precision. And pulsar-timing arrays watch binary pulsars whose spin axes precess by geodetic transport, turning distant neutron stars into gyroscopes that test the same law far outside the weak field.

The physical picture

Carry a gyroscope while you walk a winding, banked corridor. If you walked a straight line and the gyroscope felt no torque, its axis would simply be parallel-transported — every instant a faithful copy of the last. But you are turning and climbing, so your own direction of motion, the four-velocity, must turn with the corridor. Fermi-Walker transport is the instruction: at each step, rotate the gyroscope's frame by exactly the boost needed to keep your forward direction forward, and add not one degree of spatial spin beyond that. The gyroscope's axis then drifts in space for one reason only — the pure boost, applied over and over along a bending path, does not compose back to where it started. That leftover is the precession. The generator Ω^μ_ν = u^μ a_ν − a^μ u_ν is best read as the angular velocity of the observer's frame in spacetime. It lives entirely in the two-plane spanned by the four-velocity u and the four-acceleration a, and it is a pure boost there — never a spatial rotation. Because boosts in different directions do not commute, dragging the frame around a closed loop in velocity space leaves behind a rotation (the Wigner rotation of special relativity). That non-commutation, and nothing else, is the geometric origin of the precession an accelerated observer attributes to a "fixed" gyroscope.

Figure

Geodesic versus accelerated transport. On a free-fall worldline a equals zero and Fermi-Walker transport is ordinary parallel transport, so the spin arrows stay parallel. On an accelerated worldline the boost generator Ω = u∧a tilts the spin to keep it orthogonal to the turning four-velocity u, adding no spatial spin

The choice is minimal in a precise, checkable sense. Among all transport rules that preserve the metric and keep the four-velocity carried into itself, Fermi-Walker is the unique one that adds no spatial rotation; any other admissible law equals Fermi-Walker plus a chosen spatial angular velocity — the mathematical definition of "a rotating frame." So the law is not one option among many; it is the zero of rotation, the baseline against which any real gyroscope's precession is measured. And the payoff is a number for an experiment: applied to a gyroscope in low Earth orbit, the same holonomy predicts a drift of a few arcseconds per year, which Gravity Probe B set out to measure.

Figure

Geodetic precession rate of an orbiting gyroscope around Earth versus orbital radius in Earth radii, from Ω = (3/2) GM v / (c² r²) with v = √(GM/r), so Ω scales as r raised to the power −5/2. The Gravity Probe B point at r = 1.10 Earth radii gives 6.6 arcsec per year, matching the measured 6601.8 ± 18.3 mas per year

The core equations

The defining statement is a modification of parallel transport: instead of demanding that the covariant derivative along the worldline vanish, we set it equal to a specific antisymmetric operator acting on the vector. That operator is built only from the four-velocity and four-acceleration, and it is what keeps the tangent glued to the motion.

\frac{\mathrm{D}V^\mu}{\mathrm{d}\tau}\equiv u^\nu\nabla_\nu V^\mu = \big(u^\mu a_\nu - a^\mu u_\nu\big)V^\nu \qquad (1)

where: $V^\mu$ — the vector field carried along the worldline; $u^\mu$ — four-velocity, $u^\mu=\mathrm{d}x^\mu/\mathrm{d}\tau$ , normalized $u^\mu u_\mu=-1$ ; $a^\mu$ — four-acceleration, $a^\mu=u^\nu\nabla_\nu u^\mu$ ; $\tau$ — proper time; $\nabla_\nu$ — Levi-Civita covariant derivative. The right side vanishes on a geodesic, recovering parallel transport. The operator on the right is the transport generator. Lowering an index exposes its structure: it is antisymmetric, hence an infinitesimal element of the Lorentz algebra, and being built from $u\wedge a$ it is a pure boost in the velocity–acceleration plane, with no spatial rotation.

\Omega^{\mu}{}_{\nu}=u^\mu a_\nu - a^\mu u_\nu,\qquad \Omega_{\mu\nu}=u_\mu a_\nu - a_\mu u_\nu=-\Omega_{\nu\mu} \qquad (2)

where: $\Omega_{\mu\nu}$ — antisymmetric generator; the antisymmetry is what forces every inner product to be conserved along the curve. Because the generator is antisymmetric, transporting any two vectors preserves their inner product. Norms stay fixed, angles stay fixed, and an orthonormal tetrad started at one point remains orthonormal all along the worldline — the property that makes the carried frame a legitimate set of rulers and clocks.

\frac{\mathrm{d}}{\mathrm{d}\tau}\big(g_{\mu\nu}V^\mu W^\nu\big)=0 \qquad (3)

where: $W^\nu$ — a second Fermi-Walker transported vector; $g_{\mu\nu}$ — the spacetime metric, with $\nabla g=0$ (metric compatibility) doing the work behind the scenes. Two kinematic facts make the construction consistent. First, differentiating $u^\mu u_\mu=-1$ shows the four-acceleration is always orthogonal to the four-velocity — a purely spatial vector in the instantaneous rest frame.

a^\mu u_\mu=0,\qquad u^\mu u_\mu=-1 \qquad (4)

Second, and this is the entire point of the law, the four-velocity is automatically Fermi-Walker transported: feeding $u^\mu$ into the generator reproduces its own rate of change, the acceleration.

\frac{\mathrm{D}u^\mu}{\mathrm{d}\tau}=a^\mu=\Omega^{\mu}{}_{\nu}u^\nu=u^\mu(a_\nu u^\nu)-a^\mu(u_\nu u^\nu)=a^\mu \qquad (5)

where: the last equality uses $a_\nu u^\nu=0$ and $u_\nu u^\nu=-1$ . This is the check that fixes the sign of the generator and distinguishes Fermi-Walker from every other candidate. A torque-free gyroscope carried by the observer has a spin four-vector $S^\mu$ orthogonal to $u^\mu$ , and it obeys the same law. Because $S^\mu u_\mu=0$ , one of the two terms drops and the spin equation takes the compact form used to compute precession.

\frac{\mathrm{D}S^\mu}{\mathrm{d}\tau}=\big(u^\mu a_\nu - a^\mu u_\nu\big)S^\nu=u^\mu\,(a_\nu S^\nu),\qquad S^\mu u_\mu=0 \qquad (6)

where: $S^\mu$ — spin four-vector of an ideal gyroscope. In its own rest frame the spatial part of $S$ does not rotate; the drift appears only when compared to a distant inertial frame. For uniform circular motion in flat spacetime, integrating the boost over the orbit leaves a net spatial rotation per unit time — the Thomas precession, with its famous slow-motion factor of one-half.

\boldsymbol{\Omega}_{\text{Thomas}}=\frac{\gamma^2}{\gamma+1}\,\frac{\mathbf{a}\times\mathbf{v}}{c^2}\;\xrightarrow{\ v\ll c\ }\;\tfrac12\,\frac{\mathbf{a}\times\mathbf{v}}{c^2} \qquad (7)

where: $\gamma=1/\sqrt{1-v^2/c^2}$ — Lorentz factor; $\mathbf{v},\mathbf{a}$ — coordinate three-velocity and three-acceleration; $c$ — speed of light. The one-half is what corrects the spin-orbit coupling in atomic fine structure. The same holonomy in a gravitational field, applied to a gyroscope on a circular geodesic in the Schwarzschild geometry, is the geodetic (de Sitter) precession, oriented along the orbital angular momentum $\mathbf{r}\times\mathbf{v}$ .

\boldsymbol{\Omega}_{\text{geo}}=\frac{3}{2}\,\frac{GM}{c^2 r^3}\,(\mathbf{r}\times\mathbf{v}) \qquad (8)

where: $M$ — central mass; $r$ — orbital radius; the factor $3/2$ is $1$ from spatial curvature plus $1/2$ of Thomas type. Per orbit this is an angle $3\pi GM/(c^2 r)$ . Finally, a Fermi-Walker transported orthonormal triad orthogonal to $u^\mu$ builds Fermi normal coordinates: shoot spacelike geodesics off the worldline along the triad and use proper distance as the coordinate. The metric is Minkowski at the observer plus quadratic corrections carrying the acceleration (a redshift term) and the Riemann tensor (tidal terms).

g_{00}\approx-(1+a_i x^i)^2 - R_{0i0j}\,x^i x^j,\qquad g_{ij}\approx\delta_{ij}-\tfrac13 R_{ikjl}\,x^k x^l \qquad (9)

where: $x^i$ — proper spatial coordinates along the triad; $a_i$ — acceleration components; $R_{\mu i\nu j}$ — Riemann tensor at the worldline. Along the worldline ( $x^i=0$ ) the metric is exactly Minkowski, as any good local frame must be.

Derivation ADeriving the transport law from two physical requirements

The goal is to pin down, uniquely, the transport rule an accelerated observer must use for their gyroscope. We demand only two physical things — that the rule preserve every inner product (so a carried tetrad stays orthonormal and clocks stay synchronized in norm), and that it carry the four-velocity into the new four-velocity (so the observer's own direction of motion is never lost) — and we forbid any gratuitous spatial spin. We then show these three conditions leave exactly one linear transport law, the Fermi-Walker law, and that its sign is fixed by a single consistency check on $u$ .

Step 1. Posit the most general linear transport, $\mathrm{D}V^\mu/\mathrm{d}\tau=F^\mu{}_\nu V^\nu$ , and impose isometry: for all carried $V,W$ , using metric compatibility, $\mathrm{d}(g_{\mu\nu}V^\mu W^\nu)/\mathrm{d}\tau=(F_{\mu\nu}+F_{\nu\mu})V^\nu W^\mu$ . Requiring this to vanish for every pair forces $F_{\mu\nu}=-F_{\nu\mu}$ . Why: only an antisymmetric generator sits inside the Lorentz algebra, and only Lorentz transformations preserve the local Minkowski inner product — anything symmetric would stretch norms.

Step 2. Impose that the four-velocity is carried into itself. Since the actual rate of change of $u^\mu$ along the worldline is by definition the four-acceleration, we require $F^\mu{}_\nu u^\nu=\mathrm{D}u^\mu/\mathrm{d}\tau=a^\mu$ . Why: this is the promise parallel transport breaks on an accelerated curve; making it a defining condition is exactly what "repairs" parallel transport into Fermi-Walker.

Step 3. Solve the algebra. Try $F^\mu{}_\nu=u^\mu a_\nu-a^\mu u_\nu$ , which is manifestly antisymmetric. Act on $u$ : $F^\mu{}_\nu u^\nu=u^\mu(a_\nu u^\nu)-a^\mu(u_\nu u^\nu)=u^\mu(0)-a^\mu(-1)=a^\mu$ , using orthogonality and normalization. The candidate passes both requirements with coefficient one — no free constant survives. Why: had we written the opposite ordering $a^\mu u_\nu-u^\mu a_\nu$ we would have obtained $-a^\mu$ ; the sign of the generator is therefore not a convention but a consequence of $u\cdot u=-1$ .

Step 4. Establish uniqueness. Any other admissible antisymmetric $F'$ with $F'^\mu{}_\nu u^\nu=a^\mu$ differs from $F$ by a tensor $R_{\mu\nu}=-R_{\nu\mu}$ with $R_{\mu\nu}u^\nu=0$ — a generator annihilating $u$ , i.e. a purely spatial rotation. Forbidding that leftover spin singles out Fermi-Walker as the unique rotation-free choice. Why: the freedom that remains is precisely the freedom to spin your frame by hand; Fermi-Walker is the definition of not doing so, which is what a torque-free gyroscope enforces physically.

Go deeper · Derivation B · 2 minThomas precession from the boost that will not close

Derivation BThomas precession from the boost that will not close

The goal is to show that a gyroscope carried on a curved path in flat spacetime, obeying the Fermi-Walker law with a pure boost generator, nonetheless precesses spatially — and to recover the exact rate, including the notorious factor of one-half in the slow limit that fixes atomic fine structure. Working in global inertial coordinates lets covariant derivatives become ordinary ones, so the entire effect is the non-commutation of the successive boosts that keep the spin orthogonal to a turning velocity.

Step 1. In inertial coordinates the Christoffel symbols vanish, so the spin equation is $\mathrm{d}S^\mu/\mathrm{d}\tau=(u^\mu a_\nu-a^\mu u_\nu)S^\nu=u^\mu(a_\nu S^\nu)$ , using $S\cdot u=0$ . The right side is purely along $u$ , so in the momentary rest frame the spatial spin has zero rate of change instant by instant. Why: at each instant Fermi-Walker adds only a boost, never a spatial rotation — so any accumulated spatial turning must come from how the rest frames themselves stack up, not from an explicit rotation term.

Step 2. Follow the rest frame around the path. Going from the rest frame at proper time $\tau$ to that at $\tau+\mathrm{d}\tau$ is a boost; the velocity direction has also turned. Two non-collinear boosts compose into a boost times a rotation — the Wigner rotation of special relativity — and it is that residual rotation the spatial spin inherits. Why: boosts do not form a subgroup; a closed loop in velocity space returns you to the same speed but with your spatial axes rotated, which is the whole content of Thomas precession.

Step 3. Accumulate over uniform circular motion. Summing the infinitesimal Wigner rotations for velocity $\mathbf v$ turning under coordinate acceleration $\mathbf a$ gives a steady spatial angular velocity $\boldsymbol{\Omega}_{\text{Thomas}}=\dfrac{\gamma^2}{\gamma+1}\dfrac{\mathbf a\times\mathbf v}{c^2}$ , directed opposite to the orbital rotation. Why: the cross product encodes that only the component of acceleration transverse to velocity — the turning, not the speeding-up — produces precession, so straight-line acceleration gives none.

Step 4. Take the slow limit. As $\gamma\to1$ the prefactor $\gamma^2/(\gamma+1)\to1/2$ , so $\boldsymbol{\Omega}_{\text{Thomas}}\to\tfrac12\,\mathbf a\times\mathbf v/c^2$ . Why: this one-half, subtracted from the naive spin-orbit precession of an electron, is exactly what reconciles the predicted and observed hydrogen fine-structure splitting — Thomas's 1926 result.

Go deeper · Derivation C · 1 minGeodetic precession as pure parallel-transport holonomy

Derivation CGeodetic precession as pure parallel-transport holonomy

The goal is to obtain the de Sitter geodetic precession of a gyroscope on a circular geodesic orbit in the Schwarzschild field, and to see why on a geodesic the Fermi-Walker law is not needed at all — the drift is the holonomy of ordinary parallel transport around a closed spatial loop in a curved geometry. We isolate the two independent contributions that add to the celebrated coefficient three-halves.

Step 1. On a geodesic the four-acceleration is zero, so Fermi-Walker transport equals parallel transport. The gyroscope's spin is parallel-transported once around the orbit, and in a curved space it does not return to its start: it comes back rotated by a deficit angle set by the enclosed curvature. Why: this is holonomy — the same reason a vector parallel-transported around a spherical triangle returns rotated by the triangle's area times the Gaussian curvature.

Step 2. Evaluate the holonomy in the weak field. Two effects contribute. The spatial part of the Schwarzschild metric is curved, contributing a factor $1$ ; and the orbital velocity supplies a Thomas-type contribution of $1/2$ from the relativity of the moving rest frame. They add: the rate is $\boldsymbol{\Omega}_{\text{geo}}=\dfrac{3}{2}\dfrac{GM}{c^2 r^3}(\mathbf r\times\mathbf v)$ , along the orbital angular momentum. Why: separating "space curvature" from "velocity composition" explains why the coefficient is $3/2$ and not $1$ — a purely Newtonian intuition would miss the Thomas half.

Step 3. Convert to a per-orbit angle. With circular-orbit velocity $v=\sqrt{GM/r}$ and period $T=2\pi r/v$ , the accumulated precession per revolution is $\Delta\phi=\Omega_{\text{geo}}\,T=\dfrac{3\pi GM}{c^2 r}$ , a clean geometric number independent of $c$ -hidden bookkeeping. Why: writing it per orbit exposes the smallness — for Earth orbit $GM/(c^2 r)\sim10^{-9}$ , so the drift is nanoradians per revolution and must be accumulated over a year and thousands of orbits to be measured.

Limiting cases

How the single transport law specializes as the worldline and the vector change:

Case	Result	Meaning
Geodesic worldline	$a^\mu=0\Rightarrow \mathrm{D}V^\mu/\mathrm{d}\tau=0$	the generator vanishes identically; Fermi-Walker transport becomes ordinary parallel transport
The four-velocity itself	$\mathrm{D}u^\mu/\mathrm{d}\tau=\Omega^\mu{}_\nu u^\nu=a^\mu$	using $u\cdot u=-1$ , $a\cdot u=0$ ; the velocity is automatically carried into itself — the defining property
Straight-line acceleration	$\mathbf a\parallel\mathbf v\Rightarrow \boldsymbol{\Omega}_{\text{Thomas}}=0$	$\mathbf a\times\mathbf v=0$ , so a linearly accelerated (Rindler) gyroscope shows no precession, only a boost
Circular motion, flat space	$\boldsymbol{\Omega}_{\text{Thomas}}=\dfrac{\gamma^2}{\gamma+1}\dfrac{\mathbf a\times\mathbf v}{c^2}$	integrating the residual boost over the orbit gives Thomas precession; the slow limit carries the factor $1/2$
Torque-free gyroscope spin	$\mathrm{D}S^\mu/\mathrm{d}\tau=u^\mu(a_\nu S^\nu),\ S\cdot u=0$	the term $a^\mu(u_\nu S^\nu)$ drops, giving the standard gyroscope equation
Circular geodesic, Schwarzschild	$\boldsymbol{\Omega}_{\text{geo}}=\dfrac{3}{2}\dfrac{GM}{c^2 r^3}(\mathbf r\times\mathbf v)$	pure parallel-transport holonomy; $3/2 = 1$ (space curvature) $+\,1/2$ (Thomas)

Common trapFermi-Walker transport is "just a rotating frame," or conversely "just parallel transport with a twist."

Both readings are wrong, and in opposite directions. Fermi-Walker transport is not a generic rotating frame: its generator $\Omega=u\wedge a$ contains only a boost in the velocity–acceleration plane and deliberately excludes every purely spatial rotation, because a torque-free gyroscope refuses exactly that spatial spin. Add an arbitrary spatial angular velocity and you have manufactured a rotating frame, not a Fermi-Walker one — the difference is physical and measurable as extra precession. But it is also wrong to call it parallel transport with a cosmetic correction. On an accelerated worldline the two genuinely disagree, and the disagreement is the whole physics: it is what keeps the four-velocity carried into itself and what makes the carried spin precess. The two coincide if and only if the four-acceleration vanishes, i.e. on a geodesic. So the honest one-line summary is: Fermi-Walker is the unique transport that is metric-preserving, carries $u$ into $u$ , and adds zero spatial rotation — no more, no less.

Worked examples

1 · Geodetic precession of Gravity Probe B (fully numeric)

A gyroscope orbits Earth on a near-circular polar geodesic at altitude 642 km. Predict its geodetic precession rate in arcseconds per year and compare with the measured value.

Step 1. Fix the orbit. With Earth radius $R_\oplus=6.371\times10^6$ m and altitude 642 km, the orbital radius is $r=7.013\times10^6$ m $=1.10\,R_\oplus$ . Using $GM_\oplus=3.986\times10^{14}\ \mathrm{m^3/s^2}$ , the circular-orbit speed is $v=\sqrt{GM/r}=\sqrt{3.986\times10^{14}/7.013\times10^6}=\mathbf{7.54\times10^3\ m/s}$ — the familiar low-Earth-orbit speed of about 7.5 km/s.

Step 2. Evaluate the rate. With $\mathbf r\times\mathbf v$ of magnitude $rv$ for a circular orbit, $\Omega_{\text{geo}}=\dfrac{3}{2}\dfrac{GM}{c^2 r^3}\,rv=\dfrac{3}{2}\dfrac{GM v}{c^2 r^2}$ . Plugging in with $c=2.998\times10^8$ m/s: $\Omega_{\text{geo}}=1.5\times\dfrac{(3.986\times10^{14})(7.54\times10^3)}{(8.988\times10^{16})(7.013\times10^6)^2}=\mathbf{1.02\times10^{-12}\ rad/s}$ .

Step 3. Convert to observational units. Multiply by one year, $3.156\times10^7$ s, and by $206\,265$ arcsec per radian: $\Omega_{\text{geo}}\approx(1.02\times10^{-12})(3.156\times10^7)(2.063\times10^5)\approx\mathbf{6.6\ arcsec/yr}$ . The published GR prediction, refined for Earth's mass distribution, is $6606$ mas/yr; Gravity Probe B measured $\mathbf{6601.8\pm18.3\ mas/yr}$ — agreement to better than $0.3\%$ . A curvature holonomy of one part in $10^{9}$ per orbit, accumulated over a year, is a directly measured effect.

2 · Thomas precession per revolution (fully numeric)

A spinning particle moves on a uniform circular path at speed v. Show that its Thomas precession is exactly 2π(γ − 1) per revolution, then evaluate it at v = 0.6c and for the electron in the hydrogen ground state.

Step 1. Take the ratio of the Thomas rate to the orbital rate. For circular motion the coordinate acceleration is centripetal, $|\mathbf a|=v^2/r$ , and the orbital rate is $\omega_{\text{orb}}=v/r$ . Hence $\dfrac{\Omega_{\text{Thomas}}}{\omega_{\text{orb}}}=\dfrac{\gamma^2}{\gamma+1}\dfrac{v^2}{c^2}$ . Using $v^2/c^2=1-1/\gamma^2=(\gamma^2-1)/\gamma^2$ , this collapses to $\dfrac{\gamma^2}{\gamma+1}\cdot\dfrac{\gamma^2-1}{\gamma^2}=\gamma-1$ .

Step 2. Read off the per-orbit angle. Since the Thomas rate is $(\gamma-1)$ times the orbital rate, over one full revolution the spin precesses by $\Delta\phi_{\text{Thomas}}=2\pi(\gamma-1)$ , opposite in sense to the orbital motion. This compact result depends on nothing but the Lorentz factor.

Step 3. Evaluate two regimes. At $v=0.6c$ , $\gamma=1/\sqrt{1-0.36}=1.25$ , so $\Delta\phi_{\text{Thomas}}=2\pi(0.25)=\pi/2=\mathbf{90^\circ}$ per orbit — a huge, unmistakable effect for a relativistic particle. For the hydrogen ground-state electron $v=\alpha c$ with $\alpha\approx1/137$ , so $\gamma-1\approx\tfrac12\alpha^2\approx2.66\times10^{-5}$ and $\Delta\phi_{\text{Thomas}}=2\pi(2.66\times10^{-5})\approx\mathbf{1.67\times10^{-4}\ rad}$ per orbit. Small as it is, this precession halves the electron's spin-orbit coupling and is what makes the computed hydrogen fine structure match experiment.

3 · A linearly accelerated observer carries a gyroscope (contrast case)

A starship accelerates in a straight line at constant proper acceleration g = 9.8 m/s² (Rindler motion). Does a gyroscope aboard precess?

Step 1. The motion is one-dimensional, so the three-velocity $\mathbf v$ and three-acceleration $\mathbf a$ are collinear at every instant, giving $\mathbf a\times\mathbf v=0$ and hence $\boldsymbol{\Omega}_{\text{Thomas}}=\dfrac{\gamma^2}{\gamma+1}\dfrac{\mathbf a\times\mathbf v}{c^2}=0$ . There is no spatial precession at all.

Step 2. What Fermi-Walker does instead is a pure boost: the generator $\Omega=u\wedge a$ points along the motion, so a gyroscope transverse to the acceleration stays rigidly transverse and one along the motion is simply boosted with the observer. In the ship's comoving frame the spin components never change — the frame is genuinely non-rotating.

Step 3. After proper time $\tau$ the rapidity is $g\tau/c$ and $\gamma=\cosh(g\tau/c)$ . At $g=9.8\ \mathrm{m/s^2}$ , one year of proper time gives $g\tau/c=(9.8)(3.156\times10^7)/(2.998\times10^8)\approx\mathbf{1.03}$ , so $\gamma\approx1.59$ and $v\approx0.77c$ . Through this enormous change of speed the gyroscope shows zero precession. Straight-line acceleration boosts a gyroscope but never turns it; only transverse acceleration precesses it.

Physics of the portrait

Fermi-Walker transport is a rare case where the physical idea arrived from a young theorist thinking about coordinates, was completed by a pure geometer, and was made observable by a spin experiment — three threads that only later were recognized as one.

Origins · 1 minEnrico Fermi and how this was discovered

Enrico Fermi

1901–1954 · relativistic frames and the transport of spin

It is easy to forget that Enrico Fermi's first significant paper, written in 1922 while he was barely twenty-one, was about differential geometry. Asking how a small observer should set up coordinates along an arbitrary worldline, Fermi showed that one can always choose a frame in which the Christoffel symbols vanish along the entire curve — the construction now called Fermi coordinates — so that spacetime looks locally flat not just at a point but all along an observer's history, with the rule for carrying the spatial axes without spurious rotation implicit in the work. A decade later the Scottish geometer Arthur Geoffrey Walker (1909–2001), better known for the Friedmann–Lemaître–Robertson–Walker cosmologies, generalized Fermi's rule to arbitrary accelerated worldlines, giving the antisymmetric boost generator that now bears both names. Meanwhile Llewellyn Thomas, in 1926, had found the physical shadow of the same mathematics: the missing factor of one-half in the electron's spin-orbit coupling was a relativistic precession of its rest frame — the Thomas precession — which is precisely the flat-space content of Fermi-Walker transport around a curved path. Fermi went on to build the first nuclear reactor and anchor a generation of physics, but the quiet geometric insight of his student years — that an accelerated observer can carry a non-rotating frame, and that the leftover twist is physics — remains one of the cleanest bridges between kinematics and the curvature of spacetime.

History

1922

— Enrico Fermi constructs coordinates that keep the connection zero along a worldline, the seed of Fermi transport

1926

— Llewellyn Thomas explains the fine-structure factor of one-half by a relativistic precession of the electron rest frame

1932

— Arthur Geoffrey Walker generalizes Fermi's rule to arbitrary accelerated worldlines, giving Fermi-Walker transport

1959

— Leonard Schiff and George Pugh independently propose an orbiting gyroscope to measure geodetic and frame-dragging precession

2011

— Gravity Probe B reports the geodetic drift 6601.8 ± 18.3 mas per year, confirming the 3/2 GM/(c²r³) prediction

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §6.5 (Fermi-Walker transport) and §13.6 (Fermi normal coordinates) · Wald, General Relativity (1984), §4.2 and the gyroscope problems · Carroll, Spacetime and Geometry (2004), the appendix on Fermi normal coordinates · Hartle, Gravity (2003), Ch. 14 (gyroscopes and precession) · Schutz, A First Course in General Relativity (2009), Ch. 6. Two internal consistency checks anchor the signs throughout: $\Omega_{\mu\nu}=u_\mu a_\nu-a_\mu u_\nu$ is antisymmetric by inspection, and acting on $u$ gives $\Omega^\mu{}_\nu u^\nu=a^\mu$ — the sign check that distinguishes the correct generator from its negative. The Thomas rate $\gamma^2/(\gamma+1)\,\mathbf a\times\mathbf v/c^2$ carries the slow-limit factor $1/2$ , and the per-orbit form $2\pi(\gamma-1)$ follows algebraically from it. The geodetic coefficient $\tfrac32 GM/(c^2 r^3)$ gives a per-orbit angle $3\pi GM/(c^2 r)$ . The Gravity Probe B final results are: geodetic $6601.8\pm18.3$ mas/yr, frame-dragging $37.2\pm7.2$ mas/yr (Everitt et al., 2011); the $\approx6.6$ arcsec/yr estimate above uses $r=1.10\,R_\oplus$ and $GM_\oplus=3.986\times10^{14}$ m³/s².

Prerequisites. Parallel Transport · The Covariant Derivative · The Levi-Civita Connection · The Geodesic Equation · Metric Compatibility. Leads to. Tidal Forces in General Relativity · The Geodesic Deviation Equation · Affine Parameters.

Ask the tutor: Ask me to derive the Wigner rotation behind Thomas precession step by step, or to compute the geodetic and frame-dragging drifts for a gyroscope in any orbit you choose.

Is Fermi-Walker transport a new connection, or the same Levi-Civita one?

It is the same Levi-Civita connection — no new geometry is introduced. Fermi-Walker transport uses the ordinary covariant derivative $\nabla$ built from the metric; all it changes is the condition imposed along the worldline. Parallel transport sets the covariant derivative to zero; Fermi-Walker sets it equal to a specific antisymmetric operator made from the four-velocity and four-acceleration. Because that operator is built entirely from $u$ and $a$ , both of which are defined using the metric connection, nothing outside standard Riemannian geometry is needed. This is why the transport still respects metric compatibility, still preserves norms, and still reduces to parallel transport whenever the extra operator vanishes — that is, on a geodesic.

If the spin never rotates in the observer's own frame, why does it precess at all?

Because "the observer's own frame" is not a single inertial frame — it is a continuously changing sequence of momentary rest frames, each boosted relative to the last. Fermi-Walker transport guarantees the spin does not rotate spatially within each momentary rest frame. But when you compare the spin at the end of an orbit to a fixed, distant inertial frame, you have composed a whole loop of boosts, and boosts in different directions do not compose to the identity — they leave a net rotation (the Wigner rotation). That accumulated rotation is the precession. So the precession is not a torque on the gyroscope; it is the mismatch between the observer's ever-boosting frame and a global inertial standard.

How is geodetic precession possible if a gyroscope in orbit is in free fall with zero acceleration?

Exactly because it is in free fall, the analysis is clean: with $a^\mu=0$ , Fermi-Walker transport is parallel transport, and parallel transport in a curved space has holonomy. Carry a vector around any closed loop in a curved geometry and it returns rotated by an amount set by the curvature enclosed — the same reason a vector dragged around a triangle on a globe comes back turned. The orbit is such a closed spatial loop, and the Schwarzschild geometry around Earth is curved, so the spin returns each orbit slightly rotated. No force and no torque act on the gyroscope; the rotation is a property of the spacetime it is transported through.

Why does the factor one-half in Thomas precession matter for atoms?

In the electron's instantaneous rest frame, the electric field of the nucleus appears partly as a magnetic field, which torques the electron's magnetic moment and precesses its spin. Computing that precession naively — forgetting that the electron's rest frame is itself accelerating and therefore Thomas-precessing — overestimates the spin-orbit interaction by exactly a factor of two. Subtracting the Thomas precession, which is $\tfrac12\,\mathbf a\times\mathbf v/c^2$ in the slow limit, halves the coupling and brings the predicted hydrogen fine-structure splitting into agreement with experiment. Thomas found this in 1926, resolving a discrepancy that had puzzled the early quantum theorists, and it is the same one-half that appears inside the geodetic three-halves.

What exactly is the difference between a Fermi-Walker frame and a rotating frame?

A Fermi-Walker frame is the non-rotating frame of an accelerated observer: its generator contains only the boost $u\wedge a$ needed to track the velocity, and zero spatial rotation. A rotating frame is what you get by adding a spatial angular velocity $\boldsymbol{\omega}$ by hand — for instance, a frame bolted to a spinning spacecraft. Mathematically, any metric-preserving transport that carries $u$ into itself equals the Fermi-Walker law plus some antisymmetric $R_{\mu\nu}$ with $R_{\mu\nu}u^\nu=0$ ; that $R$ is precisely the extra spatial spin. Physically, a Fermi-Walker frame is realized by three torque-free gyroscopes, whereas a rotating frame would show Coriolis and centrifugal effects. The distinction is operational and measurable, not a matter of bookkeeping.

Do Fermi normal coordinates actually make gravity disappear along the worldline?

Only along the worldline, and only as far as the equivalence principle allows. On the central worldline the metric is exactly Minkowski and the Christoffel symbols vanish, so a freely falling observer sees no gravity right where they are. But move a proper distance $x^i$ off the worldline and the metric picks up quadratic corrections: a term $\sim a_i x^i$ from the observer's own acceleration and, crucially, terms $\sim R_{\mu i\nu j}x^i x^j$ from the Riemann tensor. Those curvature terms cannot be transformed away — they are the tidal forces, the true signature of gravity, and exactly what a gravitational-wave detector reads as the changing separation of its mirrors.