Propagator for the Harmonic Oscillator

The free kernel of the previous article spreads forever. Its one elementary sibling does the opposite: the harmonic oscillator's propagator refocuses — perfectly, periodically, with a phase signature so distinctive it earned its own name. This article derives the Mehler kernel,

K_\omega=\sqrt{\frac{m\omega}{2\pi i\hbar\sin\omega t}}\;\exp\!\left[\frac{im\omega}{2\hbar\sin\omega t}\Big((x^2+x'^2)\cos\omega t-2xx'\Big)\right],

and mines it for three stories. First, exactness: the exponent is again $\frac{i}{\hbar}S_{\rm cl}$ — the classical oscillator action — and the prefactor is again the Van Vleck determinant, now $\sqrt{-\frac{1}{2\pi i\hbar}\cdot\frac{-m\omega}{\sin\omega t}}$ . Quadratic Hamiltonians are the aristocracy of quantum dynamics: their semiclassical form is their exact form, a fact Van Vleck glimpsed in 1928, decades before path integrals explained why. Second, refocusing: at every half period the kernel collapses back to a delta — mirror-imaged at $\omega t=\pi$ , upright at $\omega t=2\pi$ — so every state, however contorted, reassembles exactly. No dispersion, ever; the commensurate spectrum $E_n=(n+\frac12)\hbar\omega$ makes the oscillator the one potential whose revivals cost nothing. Third, the phase ledger: each passage through a focus costs exactly $-\pi/2$ of phase — the Maslov index — and the accumulated $e^{-i\pi}$ per period is nothing else than the zero-point $\frac12$ in the level formula, read dynamically. Everything is machine-verified: the kernel solves the oscillator TDSE with zero symbolic residual, reduces to the free kernel as $\omega\to0$ , matches the Hermite spectral sum to $10^{-12}$ , and executes its mirror ( $-i\,\psi(-x)$ ) and full-period ( $-\psi$ ) refocusings numerically. The derivation route is a Gaussian ansatz with Riccati flow — no path integral required, though the doorway is pointed out.

Key takeaways

THE kernel (Mehler, 1866 mathematics): $K_\omega=\sqrt{\frac{m\omega}{2\pi i\hbar\sin\omega t}}\exp\left[\frac{im\omega((x^2+x'^2)\cos\omega t-2xx')}{2\hbar\sin\omega t}\right]$ — exact for $\hat{H}=\frac{\hat{p}^2}{2m}+\frac12m\omega^2\hat{x}^2$ .
Exact Van Vleck structure: exponent $=\frac{i}{\hbar}S_{\rm cl}$ , $\partial^2S_{\rm cl}/\partial x\partial x'=-m\omega/\sin\omega t$ , prefactor $^2=-\frac{1}{2\pi i\hbar}\partial^2_{xx'}S_{\rm cl}$ — all symbolic-verified.
$\omega\to0$ recovers the free kernel exactly (limit machine-checked); $t\to0^+$ is the δ-family.
Refocusing: $\omega t=\pi$ maps $\psi(x)\to-i\,\psi(-x)$ (mirror image, Maslov phase $-i$ ); $\omega t=2\pi$ maps $\psi\to-\psi$ (exact revival, global phase $e^{-i\pi}$ ).
The $e^{-i\pi}$ per period is the zero-point energy: $e^{-iE_0T/\hbar}=e^{-i\pi}$ for $E_0=\frac12\hbar\omega$ — Maslov phases and the $+\frac12$ are one fact.
Spectral cross-check: the Hermite bilinear sum equals the closed form (Mehler's identity), verified to $10^{-12}$ with absolute convergence at damped time.
Any Gaussian rides the classical trajectory with breathing width: $\langle x\rangle(t)=x_0\cos\omega t$ exactly (machine-checked) — no spreading, ever.
Key idea: the equally spaced harmonic-oscillator spectrum produces exact periodic revivals, with the propagator refocusing to a delta-function at integer multiples of half a period and acquiring the corresponding Maslov phase at each caustic.

The core insight

The oscillator propagator is the exponentiated classical action with its Van Vleck prefactor — exact because the Hamiltonian is quadratic — and its $\sin\omega t$ structure makes it refocus to a δ every half period, charging a Maslov phase of $-\pi/2$ per focus that, accumulated, is precisely the zero-point $\frac12\hbar\omega$ .

Where this lives today

The Mehler kernel is production machinery. Every "thermal state of a harmonic mode" computation in quantum optics and optomechanics is this kernel Wick-rotated ( $t\to-i\hbar\beta$ turns refocusing dynamics into the oscillator partition function); every Gaussian-state simulation of continuous-variable quantum information — squeezed light, cluster states, bosonic error-correcting codes — propagates covariance matrices with exactly the Riccati flow of Derivation A. Trapped-ion and tweezer-array laboratories use the refocusing theorem as an instrument: dynamical decoupling and motional-echo protocols time operations to the trap's revival clock, and departures from the exact reunion diagnose anharmonicity at the ppm level. The broader semiclassical frontier builds on the same local structure: Gutzwiller's trace formula, initial-value representations in chemical dynamics, and caustic-phase bookkeeping in wave propagation all use quadratic kernels as local models. The oscillator is especially useful because this quadratic case is exactly solvable.

The physical picture

The free kernel was a fan: momentum classes separating forever. The oscillator kernel is a lens system. Phase space rotates rigidly under $\hat{H}_{\rm QHO}$ (the operator clock of The Heisenberg Equation of Motion), so a bundle of trajectories launched from one point spreads, swings — and reconverges to a point half a period later, at the mirror position. In optics language the oscillator is a perfect relay of lenses, refocusing every $\pi/\omega$ ; the kernel's $\sin\omega t$ in the prefactor is the lens formula, diverging exactly where all trajectories cross — a caustic — and its divergence there is the kernel collapsing back to a δ.

Figure

Phase-space rotation makes the oscillator a perfect lens relay — foci every half period, caustic phase toll at each

The refocusing has a spectral explanation as clean as the geometric one. Revivals require commensurate level spacings: the oscillator's ladder $E_n=(n+\frac{1}{2})\hbar\omega$ is perfectly uniform, so all Bohr frequencies are integer multiples of $\omega$ and every relative phase resets each period. Uniquely, the reset is exact at half period too, up to a parity flip — the even/odd alternation of Hermite functions converts the half-period phase pattern $(-1)^n$ into a spatial mirror. What neither picture gives without care is the overall phase — and that is the Maslov ledger: each caustic passage multiplies the kernel by $e^{-i\pi/2}$ , so a full period costs $e^{-i\pi}$ , which is exactly $e^{-iE_0T/\hbar}$ with $E_0=\frac12\hbar\omega$ . The zero-point energy is a caustic toll.

Figure

The Maslov ledger — quarter-turn phase per focus, and the zero-point energy as accumulated toll

The kernel and its credentials

State the result, then earn it. For $\hat{H}=\hat{p}^{\,2}/2m+\frac{1}{2}m\omega^2\hat{x}^{\,2}$ and $0<t<\pi/\omega$ :

K_\omega(x,t;x',0)=\sqrt{\frac{m\omega}{2\pi i\hbar\sin\omega t}}\;\exp\!\left[\frac{im\omega}{2\hbar\sin\omega t}\Big((x^2+x'^2)\cos\omega t-2x x'\Big)\right] \qquad (1)

where: the free kernel's $1/t$ has everywhere become $\omega/\sin\omega t$ and its $(x-x')^2$ has opened into the $\cos\omega t$ -weighted quadratic form; the $\omega\to0$ limit collapses (1) back onto $K_0$ exactly (machine-checked as a symbolic limit). The credentials, all verified: (1) solves the oscillator TDSE with identically zero residual; $t\to0^+$ reproduces the δ-family; and the exponent is the classical action of the unique oscillator path from $(x',0)$ to $(x,t)$ ,

S_{\rm cl}(x,x';t)=\frac{m\omega}{2\sin\omega t}\Big[(x^2+x'^2)\cos\omega t-2xx'\Big],\qquad \frac{\partial^2S_{\rm cl}}{\partial x\,\partial x'}=-\frac{m\omega}{\sin\omega t} \qquad (2)

where: the mixed derivative again builds the prefactor, prefactor $^2=-\frac{1}{2\pi i\hbar}\partial^2_{xx'}S_{\rm cl}$ — the Van Vleck form, exact (both identities symbolic-verified). Two elementary kernels, two exact semiclassical forms: the pattern is a theorem about quadratic Hamiltonians (Gaussian integrals saturate at their stationary point), and its systematic exploitation for general Hamiltonians — where it becomes an approximation with these two kernels as local models — is the APRX chapter's program.

Derivation AThe Mehler kernel from a Gaussian ansatz

We derive (1) without path integrals, using the preservation of Gaussian form: since the Hamiltonian is quadratic, an initial Gaussian remains Gaussian, and the propagator can be obtained as the appropriate delta-function limit. We therefore posit $K=\exp\!\left[i(a(t)x^2+b(t)xx'+c(t)x'^2+d(t))/\hbar\right]$ , with $d(t)$ absorbing the normalization, and let the TDSE convert the PDE into ordinary differential equations for the coefficients. The equation for $a(t)$ is a Riccati equation — the signature of quadratic dynamics — solved by the cotangent; everything else follows by integration. The care points: matching the δ initial condition fixes all constants, and the normalization can be locked without computing a single messy integral, by demanding the composition law.

Step 1. Insert the ansatz into $i\hbar\,\partial_tK=-\frac{\hbar^2}{2m}\partial_x^2K+\frac12m\omega^2x^2K$ and match powers of $x$ and $x'$ . The $x^2$ coefficient gives the Riccati equation, the $xx'$ and $x'^2$ coefficients give linear equations slaved to it:

\dot a=-\frac{2a^2}{m}-\frac{m\omega^2}{2},\qquad \dot b=-\frac{2ab}{m},\qquad \dot c=-\frac{b^2}{2m}

Step 2. Solve the Riccati equation with the substitution $a=\frac{m}{2}\frac{\dot f}{f}$ , which linearizes it to $\ddot f=-\omega^2f$ — the classical oscillator equation, no accident. The δ initial condition ( $a\to\frac{m}{2t}$ as $t\to0$ , inherited from the free kernel's short-time form) selects $f=\sin\omega t$ :

a(t)=\frac{m\omega}{2}\cot\omega t,\qquad b(t)=-\frac{m\omega}{\sin\omega t},\qquad c(t)=\frac{m\omega}{2}\cot\omega t

Why the classical equation appears inside the quantum coefficient flow: quadratic Hamiltonians evolve Gaussian parameters by the classical symplectic flow — the same fact that made the Heisenberg operators rotate classically. The kernel's time dependence is classical mechanics wearing quantum normalization.

Step 3. Assemble the exponent: $\frac{i}{\hbar}\left[\frac{m\omega\cos\omega t}{2\sin\omega t}(x^2+x'^2)-\frac{m\omega}{\sin\omega t}xx'\right]$ — exactly $\frac{i}{\hbar}S_{\rm cl}$ of equation (2).

Step 4. Lock the prefactor by composition: demanding $K(t_1)*K(t_2)=K(t_1+t_2)$ , where $*$ denotes convolution over the intermediate position, fixes the prefactor $\sqrt{m\omega/(2\pi i\hbar\sin\omega t)}$ — the unique choice consistent with the free kernel's short-time limit, including its $e^{-i\pi/4}$ phase. Why this route generalizes: any quadratic Hamiltonian — driven oscillator, time-dependent frequency, cross terms — yields the same Gaussian-Riccati machinery; the kernel is always $e^{iS_{\rm cl}/\hbar}$ times a Van Vleck prefactor. The oscillator is the cleanest showcase, not a special trick. *The closed form's TDSE residual is identically zero (symbolic); the $t\to0^+$ and $\omega\to0$ limits check out; and the exponent equals $iS_{\rm cl}/\hbar$ while the prefactor equals the Van Vleck expression — both identities hold exactly at the symbolic level.*

Go deeper · Derivation B · 2 minRefocusing and the Maslov phase ledger

Derivation BRefocusing and the Maslov phase ledger

We derive the kernel's behavior at the caustics $\omega t=\pi$ and $2\pi$ , where $\sin\omega t\to0$ and formula (1) degenerates. The strategy is to abandon the closed form exactly there and use the spectral representation — the two forms are equal (Mehler's identity, machine-verified below), and the spectral sum is perfectly regular at the caustics. The payoff is the phase ledger: mirror refocusing with $-i$ , full revival with $-1$ , and the identification of the accumulated phase with the zero-point energy.

Step 1. Write the spectral form (The Quantum Propagator's equation, consumed) with $E_n=(n+\frac12)\hbar\omega$ :

K_\omega(x,t;x')=\sum_{n=0}^{\infty}\psi_n(x)\psi_n(x')\,e^{-i(n+\frac12)\omega t}

Step 2. Evaluate at $\omega t=\pi$ : the phases factor as $e^{-i\pi(n+\frac12)}=e^{-i\pi/2}(-1)^n$ . Use Hermite parity $(-1)^n\psi_n(x')=\psi_n(-x')$ :

K_\omega\big(x,\tfrac{\pi}{\omega};x'\big)=-i\sum_n\psi_n(x)\psi_n(-x')=-i\,\delta(x+x')

— every state reassembles mirror-imaged, carrying one quarter-turn toll: $\psi(x)\to-i\,\psi(-x)$ .

Step 3. Evaluate at $\omega t=2\pi$ : phases $e^{-2\pi i(n+\frac12)}=e^{-i\pi}$ for every $n$ — a global factor:

K_\omega\big(x,\tfrac{2\pi}{\omega};x'\big)=-\,\delta(x-x')\qquad\Longrightarrow\qquad \psi\to-\psi

— exact revival, with the full-period phase $e^{-i\pi}$ : two caustics crossed, $e^{-i\pi/2}$ each.

Step 4. Read the ledger physically. The per-period phase equals $e^{-iE_0T/\hbar}$ with $E_0=\frac12\hbar\omega$ : the Maslov toll is the zero-point energy, seen dynamically. Equivalently, the semiclassical quantization condition gives the familiar $+\frac{1}{2}$ shift: the two caustic crossings contribute a total Maslov phase of $-\pi$ , leading to $E_n=(n+\frac{1}{2})\hbar\omega$ . For the harmonic oscillator, this corresponds to Maslov index $\mu=2$ . (For $\pi<\omega t<2\pi$ , $\sin\omega t<0$ and the closed form (1) is used with the branch continued through the caustic — one extra $e^{-i\pi/2}$ ; the spectral form needs no such care, which is why it anchors this derivation.) Why this matters beyond the oscillator: every caustic in every semiclassical propagator charges the same quarter turn — the Maslov index counts foci along classical paths, and WKB's connection formulas are this ledger at a single turning point. The APRX chapter spends this systematically; the Wentzel/Brillouin story lives there. *Both refocusings hold numerically in a 60-level Hermite basis on an asymmetric test state: $\psi\to-i\psi(-x)$ at $\omega t=\pi$ to $10^{-8}$ (band-limited) and $\psi\to-\psi$ at $\omega t=2\pi$ to $10^{-10}$ ; Mehler's identity (spectral sum vs. closed form) agrees to $10^{-12}$ with absolute convergence at damped time, $N=150$ .*

What the kernel does to states

Feed (1) a displaced ground-state Gaussian and the convolution (same Gaussian calculus as the free article's Derivation B) returns a Gaussian whose center rides the classical trajectory exactly: $\langle x\rangle(t)=x_0\cos\omega t$ , machine-checked to $10^{-6}$ at three times. No approximation is involved — the exact classicality of oscillator expectation values (Ehrenfest's Theorem's linearity clause; the operator solve of The Heisenberg Equation of Motion) reappears at kernel level. A displaced ground state keeps its width rigidly (it is a coherent state — the full $\alpha$ -dynamics and its numbers are Coherent States' own); any other Gaussian breathes at $2\omega$ while its center still traces the classical cosine. And every state — Gaussian or not — obeys the same two refocusing properties: parity inversion at half a period and exact revival up to the global phase at a full period. Nothing about the oscillator disperses; the "collapse" phenomenology of anharmonic wells (the machine run of Ehrenfest's Theorem) is precisely what the uniform ladder forbids here.

Limiting cases

case	result	meaning
$\omega\to0$	$K_\omega\to K_0$ exactly	the free kernel recovered (symbolic limit)
$t\to0^+$	$K_\omega\to\delta(x-x')$	δ-family; short times don't feel the potential
$\omega t=\pi$	$-i\,\delta(x+x')$	mirror refocus + quarter-turn toll
$\omega t=2\pi$	$-\delta(x-x')$	exact revival; $e^{-i\pi}=e^{-iE_0T/\hbar}$
exponent read classically	$\frac{i}{\hbar}S_{\rm cl}$ , Van Vleck prefactor	exact — quadratic-Hamiltonian privilege
$\omega t\to\pi^-$	prefactor diverges	caustic: all classical paths converge; δ-collapse

Common trapThe kernel formula is not valid blindly across the foci

At $\omega t=n\pi$ the closed form (1) divides by zero — and the divergence is physics (a caustic: the map from launch momentum to arrival point degenerates), but using the formula there, or naively across a focus, produces wrong phases. Two honest options: switch to the spectral form, regular everywhere (how Derivation B extracts the $-i\delta(x+x')$ and $-\delta(x-x')$ results), or continue the closed form through the caustic while crediting the Maslov toll $e^{-i\pi/2}$ per crossing — for $\pi<\omega t<2\pi$ the $\sqrt{\sin\omega t}$ needs exactly that extra phase. Forgetting the toll is the classic semiclassical bug: interference fringes shift by a quarter period, and energy levels come out as $n\hbar\omega$ instead of $(n+\frac12)\hbar\omega$ — an error experiment rules out at the first spectroscopic glance.

Worked examples

Example 1 — a trapped atom's guaranteed reunion. A sodium atom sits in an optical-tweezer trap that is harmonic with $\omega/2\pi=100$ kHz. At $t=0$ the trap laser shudders, leaving the atom in some unknown contorted state $\psi_0$ . When is the state guaranteed to return, and with what signature?

Step 1. The revival clock: $T=2\pi/\omega=$ 10 μs — independent of $\psi_0$ , because the refocusing theorem is a property of the kernel, not the state.

Step 2. At $t=5\,\mu\mathrm{s}$ (half period), the state is exactly $-i\,\psi_0(-x)$ — the density is mirror-imaged, while the quarter-turn phase is invisible to a direct density measurement but appears in interference with a reference state.

Step 3. At $t=10$ μs: the state is $-\psi_0$ — literally the initial state, apart from the global Maslov phase. Any density or correlation measurement repeats exactly; only an interferometric comparison against an unshaken twin reveals the $e^{-i\pi}$ .

Step 4. Read the energy scale: $\hbar\omega=$ 0.41 neV — yet the revival is exact whether the shudder deposited one quantum or thousands, because the ladder's spacing is uniform at every rung. This is why harmonic traps are the workhorse of quantum control: dynamics in them is periodic bookkeeping, and any observed deviation from the 10 μs reunion directly measures the trap's anharmonicity.

Example 2 — certifying the kernel numerically (the Mehler audit). You implement (1) in code for a simulation. Certify it against the spectral representation, honestly.

Step 1. Naive check — sum $\sum_n\psi_n(x)\psi_n(x')e^{-i(n+\frac12)\omega t}$ at real $t$ against (1) — stalls: on the unit circle the bilinear series converges only distributionally; partial sums oscillate without settling (an instructive failure, not a bug).

Step 2. Damp the time: evaluate both sides at $t-0.15i/\omega$ (i.e. $|z|=e^{-0.15}<1$ in Mehler's variable $z=e^{-i\omega t}$ ), where the series converges absolutely. With $N=150$ terms: agreement to $10^{-12}$ at three scattered $(x,x',t)$ points — machine-run.

Step 3. Guard the Hermite evaluation: direct polynomial-coefficient formulas for $H_n$ can suffer severe loss of precision at moderately large $n$ because of cancellation; the stable route is the normalized recurrence $\psi_{n+1}=\sqrt{\frac{2}{n+1}}\,x\psi_n-\sqrt{\frac{n}{n+1}}\,\psi_{n-1}$ , which avoids this numerical instability.

Step 4. Close with the caustic test: apply the spectral kernel at $\omega t=\pi$ and $2\pi$ to an asymmetric test state and verify $-i\psi(-x)$ and $-\psi(x)$ , respectively. A kernel implementation that combines the Mehler form with the appropriate Maslov phases should remain consistent across the full time axis, including the caustics where equation (1) becomes singular.

Origins · 1 minJohn Van Vleck and how this was discovered

John Van Vleck

1899–1980 · Minneapolis, Madison, Harvard

The structure this article certifies twice — kernel $=\sqrt{-\frac{1}{2\pi i\hbar}\partial^2_{xx'}S_{\rm cl}}\;e^{iS_{\rm cl}/\hbar}$ — was written down in 1928 by the first American to earn a PhD on quantum theory, in a paper decades ahead of its audience. Van Vleck's PNAS note "The correspondence principle in the statistical interpretation of quantum mechanics" asked how Schrödinger's brand-new wave mechanics connects to classical mechanics at the amplitude level, and answered with the determinant now bearing his name: the density of classical trajectories, $\partial^2S/\partial x\partial x'$ , controls the quantum amplitude's magnitude, while the classical action controls its phase. For quadratic Hamiltonians — this article's two kernels — the formula is exact; for everything else it is the leading semiclassical approximation, a fact fully appreciated only when Feynman's path integral (1948) exposed why (stationary phase) and Gutzwiller (1967–71) built the trace formula of quantum chaos on it, Maslov indices and all. Van Vleck himself had moved on: his second act made him the architect of the quantum theory of magnetism — ligand fields, paramagnetism, the "Van Vleck susceptibility" — earning the 1977 Nobel Prize and the informal title of dean of American physics. Colleagues remembered a man who could recite railway timetables from memory and audited train routes the way this article audits kernels. The 1928 determinant, meanwhile, sits at the junction where classical action meets quantum amplitude — the precise junction the oscillator kernel exhibits with zero error.

History

1866

— Mehler publishes the bilinear Hermite identity, pure mathematics awaiting a physics

1926

— Schrödinger solves the oscillator; the ladder

(n+\tfrac12)\hbar\omega

with its curious half

1928

— Van Vleck's determinant: classical trajectory density as quantum prefactor

1948

— Feynman's path integral explains the exactness for quadratic actions

1965

— Maslov systematizes the caustic phase ledger; the index enters quantization rules

1971

— Gutzwiller's trace formula builds quantum chaos on Van Vleck propagators

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §2.6 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 8 · Feynman & Hibbs, Quantum Mechanics and Path Integrals, §3.11 · Van Vleck, PNAS 14, 178 (1928) · Schulman, Techniques and Applications of Path Integration, Ch. 6 (oscillator; caustics) · Gutzwiller, Chaos in Classical and Quantum Mechanics, Ch. 12. All equations are standard. Key numerical results: the TDSE residual is identically zero and the exponent/Van Vleck identities are exact (symbolic); the $\omega\to0$ limit is exact; Mehler's identity agrees to $10^{-12}$ (damped time $|z|<1$ , $N=150$ , stable recurrence); the caustic phases $-i\,\psi(-x)$ at $\omega t=\pi$ ( $10^{-8}$ ) and $-\psi$ at $2\pi$ ( $10^{-10}$ ) on an asymmetric state; coherent transport $\langle x\rangle=x_0\cos\omega t$ ( $10^{-6}$ ); trap numbers (10 μs, 0.41 neV) from CODATA. The Wick-rotated partition function $Z=[2\sinh(\beta\hbar\omega/2)]^{-1}$ is noted for completeness; its systematic derivation belongs to the statistical-mechanics track.

Builds on: The Quantum Propagator · Propagator for the Free Particle and Gaussian Spreading · Harmonic Oscillator (Analytic Solution) · Ladder Operators and the Algebraic Oscillator

Leads to: Green Functions in Quantum Mechanics · Semiclassical/WKB methods (APRX) · Coherent States · Path-integral methods (RQM track)

Where did Mehler's identity come from, if not physics?

Pure nineteenth-century analysis: Gustav Ferdinand Mehler proved in 1866 that $\sum_n\frac{(z/2)^n}{n!}H_n(u)H_n(v)e^{-(u^2+v^2)/2}$ sums to a closed Gaussian in $u,v$ for $|z|<1$ — a generating-function fact about Hermite polynomials. Substituting $z=e^{-i\omega t}$ (the boundary of convergence — hence the damped-time care in the audit) turns his identity verbatim into the oscillator propagator's spectral-to-closed-form bridge. It is one of the cleaner cases of mathematics stocking a shelf sixty years before quantum mechanics shopped there.

Is the −i at half period observable, or just bookkeeping?

Observable, interferometrically — like every global-phase-of-a-branch statement. A density measurement at $\omega t=\pi$ sees only the mirrored $|\psi(-x)|^2$ . But split a system into an oscillator arm and a reference arm, recombine after half a period, and the $-i$ shifts the interference fringes by a quarter period — the mechanical twin of the Gouy phase that focused light beams display, measured routinely. The full-period $-1$ is likewise visible in echo experiments; and its spectroscopic shadow is the zero-point half itself, without which every oscillator spectrum would sit $\frac12\hbar\omega$ lower.

Do the refocusing theorems survive a thermal or mixed state?

Perfectly — they are kernel properties, linear in the density operator's evolution. A mixed state $\rho$ evolves to $\hat{U}\rho\,\hat{U}^\dagger$ ; at $\omega t=\pi$ every matrix element mirrors ( $\rho(x,x')\to\rho(-x,-x')$ , the $-i$ cancelling against $+i$ from the conjugate side) and at $2\pi$ the state returns exactly ( $(-1)(-1)^*=1$ — even the Maslov phase cancels for density matrices). This is why motional-echo protocols work on thermal ions: the reunion needs no purity, only harmonicity.

What happens to the kernel — and the revivals — if the trap frequency drifts or the potential has a quartic term?

Frequency drift $\omega(t)$ keeps the Hamiltonian quadratic, so the Gaussian-Riccati machinery survives: the kernel is still exact-Van-Vleck with $f(t)$ solving the time-dependent classical equation $\ddot f=-\omega(t)^2f$ — refocusing then happens when $f$ returns to zero, generally not periodically (parametric amplification is this machinery driven on purpose). A quartic term breaks quadraticity itself: the ladder de-tunes ( $E_n$ gains an $n^2$ -ish correction), revivals become partial and then fractional, and the exact kernel is gone — the collapse phenomenology of the Ehrenfest article's machine run. Measured revival degradation is, in practice, the standard anharmonicity diagnostic.

How does the imaginary-time version give the partition function?

Set $t\to-i\hbar\beta$ in (1): $\sin\omega t\to-i\sinh(\beta\hbar\omega)$ and the kernel becomes real and positive — the thermal density matrix $\langle x|e^{-\beta\hat{H}}|x'\rangle$ . Its trace, $\int K(x,-i\hbar\beta;x)\,dx$ , is a Gaussian integral that evaluates to $[2\sinh(\beta\hbar\omega/2)]^{-1}=\sum_ne^{-\beta E_n}$ — the oscillator partition function, zero-point half included. The rotation is the germ of the entire path-integral formulation of statistical mechanics (Feynman–Kac); this track leaves its systematic use to the stat-mech shelf, flagging only that the same Mehler kernel does both jobs.

Why doesn't the oscillator kernel ever produce chaos or thermalization, when generic systems do?

Because everything about it is rigid: the spectrum is uniformly spaced (no incommensurate frequencies to dephase against forever), the classical dynamics is a pure rotation (no sensitivity to initial conditions), and the Van Vleck form is exact (no proliferation of interfering classical paths). Chaos needs anharmonicity — then classical trajectories diverge, the trace formula sums over exponentially many orbits, and revivals die into the effectively-irreversible spreading discussed at the propagator article's FAQ. The oscillator sits at the integrable extreme of that spectrum: maximal memory, zero mixing — which is exactly why laboratories build traps to be as harmonic as money allows.