Ehrenfest's Theorem

The engine is the daily language of quantum optics and optomechanics: cavity fields, mechanical oscillators and spins are tracked through coupled expectation-value equations ($\langle\hat{a}\rangle$, $\langle\hat{n}\rangle$, $\langle\hat{\sigma}_z\rangle\ldots$), and the closure question — do the equations for first moments close, or leak into higher ones? — decides whether a "semiclassical" model of a laser or a levitated nanosphere is honest. LIGO's mirror dynamics and every Kalman-filtered quantum sensor run on exactly this moment hierarchy, truncated where the $\sigma^2V'''$ audit says it may be. The failure side is equally alive: "Ehrenfest time" — the horizon beyond which a packet in a chaotic or anharmonic system abandons its classical rail (logarithmically short in $\hbar$ for chaos) — is a standard diagnostic in cold-atom experiments that launch coherent states in tilted lattices and watch, in real time, the collapse the quartic run above shows in miniature.

A wave packet is a crowd, not a runner. Each measurement of position samples the crowd once; $\langle\hat{x}\rangle$ is the crowd's centroid. Ehrenfest's question is whether the centroid runs like Newton's single particle — and the engine's answer is: the centroid feels the average of the force over the crowd, $\langle V'(x)\rangle$, not the force at the centroid, $V'(\langle x\rangle)$. Where the force is linear, the crowd average and the centroid value coincide and Newton survives verbatim. Where the force curves, the wings of the crowd vote differently from its center, and the centroid leaves the classical rail.

The second face of the engine is bookkeeping across the whole observable algebra. The Heisenberg picture rotates operators; sandwiching that rotation in any fixed state produces a coupled hierarchy of expectation-value equations — $\langle\hat{x}\rangle$ drives on $\langle\hat{p}\rangle$, $\langle\hat{p}\rangle$ on $\langle V'\rangle$, $\langle\hat{x}\hat{p}\rangle$ on $\langle 2T-xV'\rangle$, each average's motion sourced by the average one commutator away. For quadratic Hamiltonians the hierarchy closes at the first two members, which is precisely why the oscillator is exactly classical; anharmonicity leaks the motion upward into ever-higher moments, and the leak rate is the quantitative content of "the classical limit".

Everything follows from one derivation, run in the Schrödinger picture where states move and operators stand still. (The Heisenberg-register route is a one-liner by comparison: sandwich The Heisenberg Equation of Motion between fixed states. Both roads, same law — as the picture-equivalence theorem guarantees.)

where: $\langle\cdot\rangle=\langle\psi(t)|\cdot|\psi(t)\rangle$; the commutator term is the dynamics, the partial term collects built-in time dependence of the operator (usually zero); and the equation is picture-invariant — every symbol in it is a sandwich. Note the equivalent sign conventions: $\frac{1}{i\hbar}\langle[\hat{A},\hat{H}]\rangle=\frac{i}{\hbar}\langle[\hat{H},\hat{A}]\rangle$.

Two corollaries come free. The x–p pair: with $\hat{H}=\hat{p}^2/2m+V(\hat{x})$, the CCR ladder of Canonical Commutation Relation evaluates $[\hat{x},\hat{H}]=i\hbar\,\hat{p}/m$ and $[\hat{p},\hat{H}]=-i\hbar\,V'(\hat{x})$ (both machine-verified), so equation (1) reproduces $d\langle x\rangle/dt=\langle p\rangle/m$ and $d\langle p\rangle/dt=-\langle V'(x)\rangle$ — the theorems the wave chapter's Ehrenfest's Theorem derived by the integral route and owns, together with the Taylor-expansion classicality audit ($\sigma_x^2V'''/2$ apparatus) and its worked numbers. Energy: $[\hat{H},\hat{H}]=0$ gives $d\langle H\rangle/dt=\langle\partial_t\hat{H}\rangle$ — for a static Hamiltonian, the mean energy never moves, in any state whatsoever.

Here is a result the x–p pair cannot see, because it lives one rung up the moment hierarchy. Feed the engine the symmetrized product $\hat{G}=\frac{1}{2}(\hat{x}\hat{p}+\hat{p}\hat{x})$ — the quantum version of the classical virial $xp$:

where: $\hat{T}=\hat{p}^2/2m$; the commutator evaluates by the same CCR ladder, one term per factor of $\hat{G}$. In a stationary state every expectation value is frozen (Stationary States and Energy Eigenstates), so the left side vanishes and

where: for a power-law potential $V=\lambda x^n$ this reads $2\langle T\rangle=n\langle V\rangle$ — the quantum copy of the classical time-average virial theorem, with ensemble averages in place of time averages. For the oscillator ($n=2$): $\langle T\rangle=\langle V\rangle=E/2$, an exact even split, machine-checked level by level in the verifier. For Coulomb ($n=-1$): $2\langle T\rangle=-\langle V\rangle$, the energy audit that Central Potentials and Hydrogen will run on every hydrogen orbital. The theorem is a workhorse precisely because it converts one hard question (separate kinetic and potential content of an eigenstate) into one easy one (the total energy).

Bohr's correspondence principle demanded that quantum dynamics merge into classical dynamics in the appropriate limit; the engine tells us exactly when and how fast. The honest statement has two clauses. Exactness clause: if $\hat{H}$ is at most quadratic in $\hat{x}$ and $\hat{p}$ — free particle, uniform field, harmonic oscillator — then $V'$ is linear, $\langle V'(x)\rangle=V'(\langle x\rangle)$ identically, and the centroid obeys Newton exactly, for every state, however wide, however quantum. Leak clause: any curvature of the force couples the centroid to the packet's shape; expanding $V'$ about $\langle x\rangle$ (the wave article's audit) puts the leading correction at $\frac{1}{2}\sigma_x^2\,V'''(\langle x\rangle)$ — the wings of the crowd voting. The verifier runs the cleanest possible head-to-head. A displaced ground-state Gaussian ($x_0=2$, $\sigma^2=\frac12$, units $m=\omega=\hbar=1$) evolves on a fine grid, once in $V=\frac12x^2$ and once in $V=\frac12x^2+\lambda x^4$ with $\lambda=0.02$ — a mere 8% force correction at the turning point. The harmonic run tracks $x_0\cos t$ to better than $10^{-3}$ for three full periods. The quartic run agrees early (deviation 0.04 at one tenth of a period), drifts to 0.51 by the first period, peaks near 1.3, and by the third period the centroid's swing has collapsed to 1.27 against the classical 2.00 — the packet has dephased across the anharmonic frequency gradient, a miniature of the collapse Spectral Decomposition of Time Evolution anatomized. The leak is predictable: the finite width feeds an extra force $\langle 4\lambda x^3\rangle\supset12\lambda\sigma^2\langle x\rangle$, renormalizing the centroid's frequency by $\sim$6% relative to its classical twin — phase drift of exactly the observed size after one period.

Example 1 — virial audit of a trapped ion. A calcium ion sits in a harmonic trap of frequency $f=1.00$ MHz, prepared in the $n=2$ Fock state. Split its energy into kinetic and potential parts.

Step 1. The level energy: $E_2=(2+\tfrac12)\hbar\omega=2.5\,\hbar\omega$ with $\hbar\omega=\hbar\cdot2\pi\times10^6\,\mathrm{Hz}=4.14\times10^{-9}$ eV.

Step 2. So $E_2=$ 10.34 neV — a nano-electron-volt scale that makes trapped ions exquisite force sensors.

Step 3. The virial theorem for $n=2$ power law: $2\langle T\rangle=2\langle V\rangle$, i.e. $\langle T\rangle=\langle V\rangle=E_2/2=$ 5.17 neV each — an exact even split, valid level by level (machine-checked in an 80-level Fock basis), with no wavefunction integral ever computed.

Step 4. Contrast the effort: computing $\langle T\rangle$ directly means integrating $|\psi_2'(x)|^2$ against Hermite polynomials; the engine reduced it to one commutator and one division by two. That economy — total energy in, kinetic/potential split out — is why the virial theorem reappears for every hydrogen orbital in the central-potential chapter.

Example 2 — the classicality horizon of an anharmonic trap. The verifier's quartic run, read as an experiment: a packet with $\sigma^2=\frac12$ displaced to $x_0=2$ in $V=\frac12x^2+0.02\,x^4$ (dimensionless units). When does the centroid stop being classical?

Step 1. Size the perturbation: at the turning point the quartic force $4\lambda x_0^3=0.64$ is 32% of the harmonic force 2.0 — but the classical trajectory is computed in the same potential, so simple frequency shift is not the discrepancy. The quantum-classical gap comes from the width: $\langle 4\lambda x^3\rangle-4\lambda\langle x\rangle^3=12\lambda\sigma^2\langle x\rangle+\ldots=0.12\,\langle x\rangle$.

Step 2. That width term acts as a $\sim$6% frequency renormalization of the centroid against its classical twin, predicting a phase gap of $\sim0.4$ rad after one period — displacement error $\approx x_0\times0.4\times\sin(\cdot)\sim0.5$.

Step 3. The machine run confirms: deviation 0.04 at $t=T/10$ (agreement regime), 0.51 at $t=T$ (the predicted drift), maximum 1.29 by three periods.

Step 4. Beyond phase drift, the packet dephases: by the third period the quantum swing has collapsed to 1.27 vs the classical 2.00 — anharmonic level spacings running neighboring components at different clocks. The classical description fails not by small corrections but by qualitative structure (collapse, and at much longer times revival) — the Ehrenfest-time doctrine in one figure. Wave-register spreading rates and their own worked numbers are Wave Packet Spreading and Dispersion's.

References. Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §1.5 & §3.5 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §2.4 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 6 · Ehrenfest, Z. Phys. 45, 455 (1927) · Ballentine, Quantum Mechanics: A Modern Development, Ch. 14 (moment hierarchies and the classical limit).

Provenance. Derivations follow the standard treatments in Griffiths (3rd ed., §1.5, §3.5), Sakurai & Napolitano (2nd ed., §2.4), and Shankar (2nd ed., Ch. 6). Verifier should independently recompute the commutators $[x,H]=i\hbar p/m$ and $[p,H]=-i\hbar V'(x)$ from $[x,p]=i\hbar$, and the master formula by differentiating $\langle\psi|A|\psi\rangle$ with the time-dependent Schrodinger equation. The constant $\hbar=1.054571817\times10^{-34}\,\mathrm{J\,s}$ is the CODATA value. Historical attribution: Ehrenfest, Z. Phys. 45, 455 (1927). For this edition all named items were machine-executed (sympy/numpy/scipy), plus: the engine against central-difference numerics on a random 4×4 ($10^{-4}$); the virial commutator $\langle 2T-xV'\rangle=0$ in QHO eigenstates ($10^{-10}$) with $\langle T\rangle=\langle V\rangle=E/2$ exact; the trapped-ion numbers (4.14 neV, 10.34 neV, 5.17 neV); and the anharmonic head-to-head (harmonic control classical to $10^{-3}$ over 3 periods; quartic $\lambda=0.02$: deviation 0.04 at $T/10$, 0.51 at $T$, max 1.29, swing collapse 2.00→1.27). Per the chapter ownership map: the x–p pair derivations, Taylor classicality audit and its numbers belong to the wave chapter's Ehrenfest's Theorem (this article recovers the pair as corollaries and never reuses its numbers); the operator-level equation to The Heisenberg Equation of Motion; the CCR ladder to Canonical Commutation Relation (consumed); conservation theory to Constants of the Motion (next serial); spreading craft to Wave Packet Spreading and Dispersion; energy–time uncertainty explicitly NOT derived here (owned by The Energy-Time Uncertainty Relation and Mandelstam-Tamm Bound).

Builds on: Time-Dependent Schrodinger Equation · The Heisenberg Equation of Motion · Canonical Commutation Relation · Expectation Values and Variance · Stationary States and Energy Eigenstates

Leads to: Constants of the Motion and Conservation Laws · Symmetries and Conservation in Quantum Dynamics · The Quantum Propagator · Central potentials (virial audits)