Foldy-Wouthuysen Transformation

The nonrelativistic reduction of the Dirac equation, done by eliminating the small components, gives the Pauli equation and $g=2$ — but push it to the next order and it breaks: the spin-orbit coefficient comes out wrong, and the Hamiltonian loses its Hermiticity. The cure is to do the reduction right, with a genuine unitary transformation that block-diagonalizes the Dirac Hamiltonian order by order. That is the Foldy-Wouthuysen transformation, and it is the machinery that generates the entire fine structure of hydrogen — kinetic, spin-orbit (with the correct Thomas one-half), and Darwin terms — with the right coefficients and manifest Hermiticity. The Dirac Hamiltonian $H=\beta mc^{2}+c\,\boldsymbol\alpha\cdot\mathbf p+e\phi$ is a two-by-two block matrix whose off-diagonal blocks, the "odd" operator $c\,\boldsymbol\alpha\cdot\mathbf p$ , couple the particle and antiparticle sectors. Foldy and Wouthuysen found the rotation that removes this coupling. For a free particle the rotation is exact — a single momentum-dependent boost angle $\tan 2\theta=|\mathbf p|/mc$ turns $H$ into $\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ , the relativistic energy with the sectors fully decoupled. With external fields the potential re-mixes the blocks, and the decoupling becomes an order-by-order expansion in $1/mc^{2}$ — an asymptotic series rather than a closed form. This article owns the transformation itself: the exact free-particle result, the iterative construction with fields, and the mechanism by which the fine-structure operators fall out of the algebra — including the automatic Thomas factor that the naive reduction misses. Its most vivid physical byproduct is Zitterbewegung, the trembling motion of the Dirac position operator at frequency $2mc^{2}/\hbar$ , which the FW frame removes to give a smooth mean position. The assembly of the three fine-structure operators into the hydrogen spectrum is developed in Relativistic Corrections to the Hydrogen Atom by name.

Key takeaways

The Dirac Hamiltonian $H=\beta mc^{2}+\mathcal O+\mathcal E$ has an "odd" part $\mathcal O=c\,\boldsymbol\alpha\cdot\boldsymbol\pi$ that couples particle and antiparticle sectors; FW is the unitary $e^{iS}$ that removes it.
Free particle (exact, owned): $U=\exp(\beta\,\boldsymbol\alpha\cdot\hat{\mathbf p}\,\theta)$ with $\tan 2\theta=|\mathbf p|/mc$ gives $H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ — the series terminates, sectors fully decoupled.
With fields (owned): the decoupling is iterative in $1/mc^{2}$ ; each step demotes the odd operator by one power of $1/m$ , generating even corrections from commutators $[\mathcal O,[\mathcal O,\mathcal E]]$ .
Punchline: the spin-orbit and Darwin terms fall out with the correct coefficients — including the Thomas one-half the naive reduction gets wrong — and the Hamiltonian stays Hermitian.
Zitterbewegung: the Dirac velocity $c\boldsymbol\alpha$ oscillates at $\omega_Z=2mc^{2}/\hbar\approx1.55\times10^{21}\ \text{s}^{-1}$ over the Compton length $\hbar/mc\approx386\ \text{fm}$ ; FW removes it, giving a smooth mean position.
Key idea: FW is the unitary block-diagonalization of the Dirac Hamiltonian — exact for the free particle ( $\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ ), order-by-order with fields — and it generates the full fine structure with the automatic Thomas half.

The core insight

The Foldy-Wouthuysen transformation is the unitary rotation $e^{iS}$ that block-diagonalizes the Dirac Hamiltonian, decoupling positive- and negative-energy sectors. For a free particle it is exact, with angle $\tan 2\theta=|\mathbf p|/mc$ giving $H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ ; with external fields it is an order-by-order expansion in $1/mc^{2}$ that generates the kinetic, spin-orbit (Thomas half automatic), and Darwin corrections. It also removes Zitterbewegung, replacing the trembling Dirac coordinate with a smooth mean position.

Where this lives today

The Foldy-Wouthuysen transformation is the theoretical engine of relativistic quantum chemistry. Computing the electronic structure of heavy atoms and molecules — where the inner electrons move at an appreciable fraction of $c$ — requires relativistic corrections, and the modern Douglas–Kroll–Hess methods that deliver them are direct, higher-order descendants of the FW block-diagonalization. Gold's color, mercury's liquidity at room temperature, and the voltage of the lead-acid battery are all relativistic effects that these methods capture. In atomic physics the FW-derived fine-structure operators are the starting point for precision spectroscopy and for the extraction of fundamental constants from atomic energy levels. And Zitterbewegung, long a formal curiosity, has been simulated in the laboratory: trapped ions and cold-atom systems can be engineered to obey a Dirac-like equation with an artificially small $mc^{2}$ , slowing the trembling from $10^{21}\ \text{Hz}$ to observable kilohertz frequencies, so that the oscillation the FW transformation was invented to remove can now be watched directly.

Figure

Zitterbewegung: the Dirac coordinate trembles at 2mc²/ℏ; the FW mean position is smooth

The physical picture

Think of the Dirac Hamiltonian as a block matrix whose off-diagonal blocks — the odd operator $\mathcal O=c\,\boldsymbol\alpha\cdot\mathbf p$ — couple the electron sector to the antiparticle sector. FW is the change of basis that rotates this matrix to (nearly) block-diagonal form. For the free particle the rotation angle depends on momentum, $\tan 2\theta=|\mathbf p|/mc$ , and it is exactly the "boost" that aligns the spinor with its own energy eigenbasis. The result is that the transformed Hamiltonian is simply $\beta$ times the relativistic energy — no coupling left, no odd remainder. Treat the odd operator $\mathcal O$ as the "error" that mixes the blocks. Each FW step is a similarity transform $e^{iS}He^{-iS}$ with $S\sim\mathcal O/m$ tuned so that the first-order term cancels $\mathcal O$ . What survives are commutators $[\mathcal O,\mathcal E]$ and $\mathcal O^{2}/m$ , smaller by a factor $1/m$ — these are the corrections. Iterate, and the odd part shrinks by one power of $1/m$ each time, while the even part accumulates the fine-structure operators. The spin-orbit and Darwin terms are not put in by hand: they fall out of the algebra of $[\boldsymbol\alpha\cdot\mathbf p,V]$ , with the correct relativistic coefficients, Thomas half included.

Figure

FW rotates the Dirac Hamiltonian to block-diagonal form, decoupling particle and antiparticle

The transformation

We build the FW transformation in two stages: first the exact free-particle rotation, then the iterative construction with fields that generates the corrections.

Derivation AThe exact free-particle Foldy-Wouthuysen transformation

We construct the unitary $U=\exp(\beta\,\boldsymbol\alpha\cdot\hat{\mathbf p}\,\theta)$ and show that a single momentum-dependent angle diagonalizes the free Dirac Hamiltonian exactly, giving $H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ . The strategy is that for the free particle the only odd operator is $c\boldsymbol{\alpha}\cdot\mathbf{p}$ , which anticommutes with $\beta mc^2$ , so one rotation in the plane spanned by $\beta mc^{2}$ and $c\,\boldsymbol\alpha\cdot\mathbf p$ can cancel the odd part completely — like diagonalizing a $2\times2$ real symmetric matrix.

Step 1. The generator squares to $-1$ . With the free Hamiltonian $H=c\,\boldsymbol\alpha\cdot\mathbf p+\beta mc^{2}$ , take momentum along $\hat z$ and the generator direction $K=\beta\alpha_z$ . Using $\{\beta,\alpha_z\}=0$ and $\alpha_z^{2}=1$ ,

(\beta\alpha_z)^{2}=\beta\alpha_z\beta\alpha_z=-\beta^{2}\alpha_z^{2}=-\mathbb 1. \qquad (1)

Why this matters: $K^{2}=-1$ means $e^{\theta K}=\cos\theta+K\sin\theta$ — the rotation is a clean trigonometric object, not an infinite series.

Step 2. Rotate. With $U=\cos\theta+\beta\alpha_{\mathbf p}\sin\theta$ , where $\alpha_{\mathbf p}=\boldsymbol{\alpha}\cdot\hat{\mathbf p}$ , which for the momentum along $\hat z$ chosen in Step 1 is just $\alpha_z$ , conjugating $H$ mixes the $\beta$ and $\alpha_{\mathbf p}=\boldsymbol\alpha\cdot\hat{\mathbf p}$ (here $=\alpha_z$ , since $\mathbf p$ was taken along $\hat z$ ) pieces just as a rotation mixes two axes:

UHU^{-1}=\beta\bigl(mc^{2}\cos2\theta+cp\sin2\theta\bigr)+\alpha_z\bigl(cp\cos2\theta-mc^{2}\sin2\theta\bigr). \qquad (2)

where: the double-angle appears because the generator acts once from each side. The second bracket is the surviving odd part.

Step 3. Kill the odd part. Demand the $\alpha_z$ coefficient vanish:

cp\cos2\theta-mc^{2}\sin2\theta=0\;\Longrightarrow\;\tan 2\theta=\frac{cp}{mc^{2}}=\frac{|\mathbf p|}{mc}. \qquad (3)

Why this is the "boost": $\tan2\theta=|\mathbf p|/mc$ is exactly the rapidity-like angle that aligns the spinor with the energy eigenbasis; it depends on momentum, which is why the transformation is nonlocal.

Step 4. Read the energy. Substituting (3) into the $\beta$ coefficient,

H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}. \qquad (4)

The sectors are fully decoupled: the upper block is $+\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ , the lower $-\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ , with no odd remainder. Why exact: with no potential, nothing regenerates the odd operator, so one rotation finishes the job — the FW series terminates. *Three quick checks confirm the result: $(\beta\alpha_z)^{2}=-\mathbb 1$ (the generator really squares to $-1$ ); at $\tan 2\theta=|\mathbf p|/mc$ the odd (alpha) block vanishes and the even block is exactly $\pm\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ ; and at the concrete point $p=mc$ the formula gives $E=\sqrt2\,mc^{2}$ , consistent with the worked example above.*

Go deeper · Derivation B · 1 minIterative FW with fields, and the automatic Thomas half

Derivation BIterative FW with fields, and the automatic Thomas half

We show how one FW step demotes the odd operator by a power of $1/m$ and generates the even corrections, and why the spin-orbit term comes out with the Thomas factor of one-half automatically. The strategy: choose the generator to cancel the leading odd term; the Baker–Campbell–Hausdorff expansion then leaves commutators that are the fine-structure operators. Write $H=\beta mc^{2}+\mathcal O+\mathcal E$ with $\mathcal O=c\,\boldsymbol\alpha\cdot\boldsymbol\pi$ (odd, $\{\mathcal O,\beta\}=0$ ) and $\mathcal E=e\phi$ (even, $[\mathcal E,\beta]=0$ ).

Step 1. Expand. For static fields, $H'=e^{iS}He^{-iS}=H+i[S,H]+\frac{i^{2}}{2}[S,[S,H]]+\cdots$ , keeping the terms needed through the order $1/m^{3}$ shown below.

Step 2. Choose the generator. Take

S=-\frac{i\beta\mathcal O}{2mc^{2}}, \qquad (5)

tuned so the $O(\mathcal O)$ part of $H'$ cancels. Why this $S$ : $i[S,\beta mc^{2}]=-\mathcal O$ exactly cancels the leading odd term, leaving a new odd part smaller by $1/m$ .

Step 3. Collect the even terms. The BCH expansion yields

H'=\beta mc^{2}+\mathcal E+\frac{\beta\mathcal O^{2}}{2mc^{2}}-\frac{1}{8m^{2}c^{4}}\bigl[\mathcal O,[\mathcal O,\mathcal E]\bigr]-\frac{\beta\mathcal O^{4}}{8m^{3}c^{6}}+\mathcal O', \qquad (6)

with $\mathcal O'=O(1/m)$ the residual odd part (removed by a second transformation). Why these terms: $\beta\mathcal O^{2}/2mc^{2}$ contains the nonrelativistic kinetic energy and the Zeeman term; $\mathcal O^{4}$ gives the $-p^{4}/8m^{3}c^{2}$ relativistic correction; the double commutator $[\mathcal O,[\mathcal O,\mathcal E]]$ is where spin-orbit and Darwin live.

Step 4. Extract spin-orbit with the Thomas half. Evaluating the double commutator for a central potential $\mathcal E=V(r)$ , using $\mathcal O=c\,\boldsymbol\alpha\cdot\mathbf p$ and $[\mathbf p,V]=-i\hbar\nabla V$ , the spin-dependent piece is

H_{\text{so}}=\frac{1}{2m^{2}c^{2}}\frac{1}{r}\frac{dV}{dr}\,\mathbf S\cdot\mathbf L. \qquad (7)

where: the coefficient carries a factor of $1/2$ relative to the naive rest-frame magnetic-field estimate. The naive argument gives $(1/m^2c^2)(1/r)(dV/dr)\,\mathbf S\cdot\mathbf L$ , which is twice too large. The missing factor of $1/2$ is the Thomas-precession correction, a relativistic effect arising from successive non-commuting boosts. In FW it emerges automatically from the same double commutator, so no separate precession argument is needed. Two checks on the iterative result: the free-particle energy $\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ expands as $mc^{2}+p^{2}/2m-p^{4}/8m^{3}c^{2}$ with coefficients $1,\tfrac12,-\tfrac18$ ; and the FW spin-orbit coefficient is exactly half the naive rest-frame estimate — the Thomas factor emerges automatically from the double commutator.

Limiting cases

regime	FW Hamiltonian	reading
free particle (exact)	$\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$	series terminates; sectors fully decoupled
leading nonrelativistic	$mc^{2}+p^{2}/2m+V$	recovers Schrödinger/Pauli; matches naive elimination
order $1/m^{2}c^{2}$	$-p^{4}/8m^{3}c^{2}+H_{\text{so}}+H_{\text{Darwin}}$	the full fine structure, Thomas half included
strong field / relativistic	series diverges	the $1/m$ expansion is asymptotic and becomes unreliable when relativistic effects are large; use the exact Dirac equation
Zitterbewegung	removed in the FW frame	the trembling lives in the odd part; the mean position is smooth

Common trapThe FW series with fields is asymptotic, not convergent — and only free-particle FW is exact

It is easy to over-read the exact free-particle result and imagine that FW always gives a closed-form, exactly decoupled Hamiltonian. It does not. Only the free transformation terminates; with external fields each step generates new odd terms through $[\mathcal O,\mathcal E]$ , so the decoupling is an order-by-order expansion in $1/mc^{2}$ that is asymptotic — it improves for a while and then diverges. For light atoms ( $Z\alpha\ll1$ ) truncating at order $1/m^{2}$ gives the fine structure to superb accuracy, but for heavy, highly-charged ions the expansion fails and you must solve the Dirac equation exactly (developed in Exact Dirac Solution of the Hydrogen Atom). The FW Hamiltonian is a controlled approximation in a small parameter, not an exact rewriting of the interacting theory.

Worked examples

Example 1 — the free FW energy and its expansion (confidence builder). Verify the exact free FW result at a concrete momentum and recover the relativistic kinetic correction.

Step 1. Exact energy at $p=mc$ . From $H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ , at $|\mathbf p|=mc$ ,

E=\sqrt{(mc)^{2}c^{2}+m^{2}c^{4}}=\sqrt{2}\,mc^{2}.

For an electron ( $mc^2=0.511\,\mathrm{MeV}$ ), this gives $E\approx0.723\,\mathrm{MeV}$ ; the corresponding speed is $v/c=1/\sqrt{2}\approx0.71$ , so the electron is already in the relativistic regime.

Step 2. Expand for small momentum. Writing $E=mc^{2}\sqrt{1+(p/mc)^{2}}$ and expanding,

E=mc^{2}+\frac{p^{2}}{2m}-\frac{p^{4}}{8m^{3}c^{2}}+\cdots

The coefficients $1,\tfrac12,-\tfrac18$ are machine-verified; the first is rest energy, the second the Newtonian kinetic energy, the third the leading relativistic correction.

Step 3. Identify the correction. The term $-p^{4}/8m^{3}c^{2}$ is exactly the relativistic kinetic energy correction, developed in Relativistic Kinetic Energy Correction — here derived as the $\mathcal O^{4}$ term of the FW Hamiltonian rather than by expanding $\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ by hand. Its size in hydrogen is $\sim\alpha^{2}$ times the gross-structure energy, of order $10^{-4}\ \text{eV}$ (recomputed).

Example 2 — the Zitterbewegung frequency and amplitude (exam level). Compute the trembling frequency of the Dirac position operator and the length scale it smears over.

Step 1. The velocity oscillates. In the Heisenberg picture the Dirac velocity operator is $\dot{\mathbf x}=(i/\hbar)[H,\mathbf x]=c\boldsymbol\alpha$ , whose eigenvalues are $\pm c$ — a sign that $\mathbf x$ is not a smooth particle coordinate. Its equation of motion integrates to an oscillation at angular frequency $2H/\hbar$ .

Step 2. The frequency. For a nearly-at-rest electron $H\approx mc^{2}$ , so

\omega_Z=\frac{2mc^{2}}{\hbar}=\frac{2(0.511\ \text{MeV})}{\hbar}\approx\mathbf{1.55\times10^{21}\ \text{s}^{-1}}.

This is a colossal frequency — orders of magnitude faster than any atomic timescale (recomputed).

Step 3. The amplitude. The characteristic length scale of the oscillation is set by the reduced Compton wavelength,

\lambda_C=\frac{\hbar}{mc}\approx\mathbf{3.86\times10^{-13}\ \text{m}}=386\ \text{fm}.

So the Dirac electron "trembles" over a Compton-scale region at an angular frequency $\omega_Z\sim10^{21}\ \text{s}^{-1}$ . The FW transformation replaces this jittery coordinate with a smooth mean position $\mathbf X_{\text{FW}}$ whose velocity is the sensible $c^{2}\mathbf p/E$ . Zitterbewegung is the physical face of the sector coupling FW removes, and the Darwin term (developed in The Darwin Term and Zitterbewegung) is its residue in bound states (recomputed).

From a trembling electron to a smooth one

Origins · 1 minfoldy-wouthuysen transformation and how this was discovered

foldy-wouthuysen transformation

Leslie Foldy 1919–2001, Siegfried Wouthuysen 1923–2001 · Cleveland, Amsterdam, 1950

The transformation is one of the rare pieces of physics named for two collaborators who solved a specific technical problem cleanly. Leslie Foldy and Siegfried Wouthuysen, working at Case Institute in Cleveland, published their systematic unitary decoupling of the Dirac equation in Physical Review in 1950, giving for the first time a controlled, Hermitian, order-by-order route from the four-component relativistic theory to a two-component particle Hamiltonian. Before them, the reduction was done by the ad hoc elimination of small components, which produced non-Hermitian operators and, notoriously, the wrong spin-orbit coefficient — the missing Thomas half. Foldy and Wouthuysen's method made the Thomas factor automatic: it emerges from the same double commutator that gives spin-orbit, with no need for the separate 1926 kinematic argument of Llewellyn Thomas. Their paper also clarified the meaning of the position operator in relativistic quantum mechanics, distinguishing the jittery Dirac coordinate — the seat of Erwin Schrödinger's 1930 Zitterbewegung — from the smooth "mean position" that a physical electron actually follows. The FW transformation remains the standard bridge between the Dirac equation and atomic and molecular structure, and its modern descendants (Douglas–Kroll–Hess) are the workhorses of relativistic quantum chemistry.

History

1926

— Thomas explains the factor-of-two discrepancy in spin-orbit via relativistic precession

1930

— Schrödinger identifies Zitterbewegung in the free Dirac equation

1948

— Pryce and Tani analyze the relativistic position and mean-position operators

1950

— Foldy and Wouthuysen publish the systematic unitary transformation, Phys. Rev. 78, 29

1974

— Douglas and Kroll extend the method for relativistic quantum chemistry

Going further

References. Foldy & Wouthuysen, Phys. Rev. 78, 29 (1950) · Bjorken & Drell, Relativistic Quantum Mechanics, §4.3 · Sakurai, Advanced Quantum Mechanics, §3.3 · Greiner, Relativistic Quantum Mechanics, Ch. 11 · Schrödinger, Sitzungsber. Preuss. Akad. Wiss. (1930) (Zitterbewegung). The key results derived in this article are: (1) the exact free-particle transformation $U=\exp(\beta\,\boldsymbol\alpha\cdot\hat{\mathbf p}\,\theta)$ with $\tan 2\theta=|\mathbf p|/mc$ , giving $H'=\beta\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ — confirmed by $(\beta\alpha_z)^{2}=-\mathbb 1$ , the vanishing of the odd part at the FW angle, and even blocks $\pm\sqrt{p^{2}c^{2}+m^{2}c^{4}}$ ; (2) the iterative construction with generator $S=-i\beta\mathcal O/2mc^{2}$ , demoting the odd part by $1/m$ per step and generating the even corrections; (3) the automatic Thomas half in the spin-orbit coefficient from the double commutator; and (4) Zitterbewegung at $\omega_Z=2mc^{2}/\hbar\approx1.55\times10^{21}\ \text{s}^{-1}$ over $\lambda_C=\hbar/mc\approx386\ \text{fm}$ , together with the $p^{4}$ expansion with coefficients $1,\tfrac12,-\tfrac18$ . The three fine-structure operators are assembled for hydrogen in Relativistic Corrections to the Hydrogen Atom; the exact treatment is Exact Dirac Solution of the Hydrogen Atom; the Darwin term and its Zitterbewegung origin are developed in The Darwin Term and Zitterbewegung.

Builds on: The Dirac Equation · Dirac Gamma Matrices and Clifford Algebra · Dirac Spinors and Plane-Wave Solutions · Nonrelativistic Reduction and the Pauli Equation

Leads to: Relativistic Corrections to the Hydrogen Atom · Spin-Orbit Coupling from Relativity · The Darwin Term and Zitterbewegung

Why can the free-particle FW transformation be done exactly while the interacting one cannot?

For the free particle the only odd operator is $c\,\boldsymbol\alpha\cdot\mathbf p$ , and it commutes with itself and combines with $\beta mc^{2}$ in a way that lets a single momentum-dependent rotation cancel it exactly, with $\tan 2\theta=|\mathbf p|/mc$ . Nothing regenerates the odd part, so the series terminates. With external fields the potential $e\phi$ fails to commute with $\boldsymbol\alpha\cdot\boldsymbol\pi$ , so each transformation generates new odd terms through $[\mathcal O,\mathcal E]$ . These can only be removed iteratively, giving an asymptotic $1/m$ series rather than a closed form. The free case is special precisely because there is nothing to spoil the single clean rotation.

How does FW automatically supply the Thomas factor of one-half in spin-orbit?

The spin-orbit term arises from the double commutator $-\tfrac{1}{8m^{2}c^{4}}[\mathcal O,[\mathcal O,\mathcal E]]$ in the transformed Hamiltonian. Evaluating it for a central potential produces $\tfrac{1}{2m^{2}c^{2}}\tfrac1r\tfrac{dV}{dr}\mathbf S\cdot\mathbf L$ — with the factor $\tfrac12$ already present. The naive rest-frame argument (transform to the electron's instantaneous frame, compute the magnetic field it sees) gives twice this. The missing half is Thomas precession, a purely relativistic kinematic effect of the fact that successive Lorentz boosts do not commute. In FW you never invoke Thomas precession separately: the Dirac equation already encodes it, and the double commutator delivers the correct coefficient. That is one of the strongest selling points of the method over the naive elimination.

What exactly is Zitterbewegung, and is it real?

Zitterbewegung is the rapid oscillation of the Dirac position and velocity operators. The velocity operator $c\boldsymbol\alpha$ has eigenvalues $\pm c$ and does not commute with the Hamiltonian, so a localized wave packet's centroid trembles at frequency $\omega_Z=2mc^{2}/\hbar\approx1.55\times10^{21}\ \text{s}^{-1}$ over an amplitude of the Compton wavelength $\hbar/mc\approx386\ \text{fm}$ . Physically it is the interference between the positive- and negative-energy components that any localized state must contain. For a free electron it is not directly observable — you cannot localize an electron to within a Compton wavelength without pair creation — but it has been simulated in trapped-ion and cold-atom systems tuned to obey a Dirac-like equation with a small effective mass, where the trembling slows to observable frequencies.

Is the FW-transformed Hamiltonian exact, or an approximation?

For a free particle it is exact. With external fields it is an approximation: an order-by-order expansion in $1/mc^{2}$ that is asymptotic, meaning it gives excellent results when truncated at low order for weak binding ( $Z\alpha\ll1$ ) but eventually diverges if pushed too far. Truncating at order $1/m^{2}$ gives the fine structure of light atoms to high accuracy. For heavy, highly-charged ions where $Z\alpha$ approaches one, the expansion breaks down and you must solve the Dirac equation exactly. So the FW Hamiltonian is a controlled effective Hamiltonian, not an exact rewriting of the interacting theory.

How is FW different from just eliminating the small components?

Both aim to reduce four components to two, but the naive elimination is not a unitary transformation, so it does not preserve the inner product exactly. This has two consequences: the resulting Hamiltonian is not Hermitian beyond leading order, and specific coefficients — most famously the spin-orbit Thomas half — come out wrong. FW replaces the elimination with a sequence of genuine unitary rotations $e^{iS}$ that block-diagonalize the Hamiltonian order by order. The transformed Hamiltonian is manifestly Hermitian, the states stay normalized, and every fine-structure coefficient is correct. FW is the disciplined, systematic version of the reduction; the naive elimination is a shortcut that works only at leading order.

Why does the FW position operator differ from the Dirac position x?

Because the Dirac position operator $\mathbf x$ is not a good particle coordinate: its velocity $c\boldsymbol\alpha$ has eigenvalues $\pm c$ and trembles (Zitterbewegung), reflecting that $\mathbf x$ mixes the positive- and negative-energy sectors. The FW transformation, by decoupling those sectors, defines a new "mean position" operator $\mathbf X_{\text{FW}}=U\mathbf x U^{-1}$ whose velocity is the sensible group velocity $c^{2}\mathbf p/E$ with no trembling. $\mathbf X_{\text{FW}}$ is the coordinate that behaves like a classical particle position and that appears in the nonrelativistic reduction; the Darwin term in atoms is precisely the correction that accounts for the difference between $\mathbf x$ and $\mathbf X_{\text{FW}}$ in a potential.