Equivalence of the Dynamical Pictures

Three pictures now run in parallel. Schrödinger moves states, Heisenberg moves operators, Dirac's interaction picture splits the difference — and the chapter has been promising, at each step, that nothing measurable could ever tell them apart. The Schrodinger Picture stated the invariances; The Heisenberg Picture kept its ledger of who carries time; the previous article leaned on register-matching without proof. This article pays the debt: it proves the equivalence theorem, and it proves something better — that the three pictures are merely three entries in an infinite catalogue. The master construction takes any time-dependent unitary $\hat{W}(t)$ and re-drapes the theory: states become $\hat{W}|\psi\rangle$ , operators become $\hat{W}\hat{A}\hat{W}^\dagger$ , and the state's new equation of motion is governed by an effective Hamiltonian $\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\dot{\hat{W}}\hat{W}^\dagger$ . That second term — the price of a moving frame, quantum mechanics' version of a fictitious force — is the whole story. Choose $\hat{W}=\hat{1}$ and you have Schrödinger. Choose $\hat{W}=\hat{U}^\dagger$ and the effective Hamiltonian cancels to zero: states freeze, and the Heisenberg picture is no longer a postulated alternative but a corollary. Choose $\hat{W}=e^{i\hat{H}_0t/\hbar}$ and out falls $\hat{H}_W=\hat{V}_I$ — the interaction picture, derived rather than declared. Every choice leaves matrix elements, spectra and Born probabilities untouched, and the proof is three lines of cancelling unitaries. The physics is the invariant core; the pictures are coordinates. Eckart glimpsed this in 1926, proving wave and matrix mechanics were one theory in two dressings; the rotating frames running today's qubit gates are the same theorem cashed as engineering.

Key takeaways

Any unitary $\hat{W}(t)$ defines a legitimate picture: $|\psi_W\rangle=\hat{W}|\psi_S\rangle$ , $\hat{A}_W=\hat{W}\hat{A}_S\hat{W}^\dagger$ — infinitely many pictures, three of them famous.
The transformed state obeys $i\hbar\,\partial_t|\psi_W\rangle=\hat{H}_W|\psi_W\rangle$ with the master formula $\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\,\dot{\hat{W}}\hat{W}^\dagger$ .
$\hat{W}=\hat{1}$ : Schrödinger. $\hat{W}=\hat{U}^\dagger$ : $\hat{H}_W=0$ — the Heisenberg picture derived. $\hat{W}=e^{i\hat{H}_0t/\hbar}$ : $\hat{H}_W=\hat{V}_I$ — the interaction picture derived.
Invariants of every picture: inner products and matrix elements $\langle\varphi|\hat{A}|\psi\rangle$ , spectra of observables, Born probabilities (with base kets transformed consistently), expectation values.
The extra term $i\hbar\dot{\hat{W}}\hat{W}^\dagger$ is the quantum fictitious force — in a rotating spin frame it is the $-\hbar\omega\hat{\sigma}_z/2$ that turns a driven Hamiltonian static.
Physics = the invariants; a picture = a choice of coordinates on Hilbert space. Choose the frame that makes your problem static.
Key idea: all pictures are unitary re-drapings of one dynamics; the effective Hamiltonian transforms with an extra $i\hbar\dot{W}W^\dagger$ , and every measurable number is frame-invariant.

The core insight

Dynamical pictures are time-dependent unitary changes of frame: $|\psi_W\rangle=\hat{W}|\psi\rangle$ , $\hat{A}_W=\hat{W}\hat{A}\hat{W}^\dagger$ , $\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\dot{\hat{W}}\hat{W}^\dagger$ — and all matrix elements, spectra and probabilities are invariant under every such change.

Where this lives today

Every quantum-control laboratory runs on engineered pictures. Superconducting-qubit gates are designed entirely in rotating frames — Example 1 is the literal first page of a transmon control derivation — and modern pulse-optimization stacks chain several frames (drive frame, then interaction frame of the residual, then an averaging frame), each one an application of the master formula with a cleverer $\hat{W}$ . Floquet engineering runs the logic in reverse: choose the frame term you want, then build a drive whose $i\hbar\dot{\hat{W}}\hat{W}^\dagger$ synthesizes an effective Hamiltonian nature didn't provide — artificial gauge fields and topological band structures in cold atoms are manufactured fictitious forces. The theorem also polices theory. Any calculated number that differs between pictures is wrong by construction — a standard debugging tool in quantum-optics and QFT computations, where "check picture independence" catches misplaced frame factors the way dimensional analysis catches unit slips.

The physical picture

A skater spins in an ice rink. Describe her from the stands, from a camera turning with her, or from a drone circling at some third rate: the coordinates differ wildly, but her blade's bite into the ice — every contact, every measurable — is identical in all three accounts. Hilbert space supports exactly this freedom. A state and its observables can be jointly rotated by any time-dependent unitary without touching the numbers experiments produce, because every measurable quantity is built from sandwiches $\langle\varphi|\hat{A}|\psi\rangle$ in which the frame factors meet their daggers and vanish.

Figure

One dynamics, a dial of frames — W(t) slides the time burden between states and operators

What changes between frames is the effective Hamiltonian — and it does not change the way ordinary operators do. Conjugation $\hat{W}\hat{H}\hat{W}^\dagger$ is only half its transformation; a moving frame adds a genuinely new piece, $i\hbar\dot{\hat{W}}\hat{W}^\dagger$ , that has no analogue for other observables. Classical mechanics students have met this creature: step onto a merry-go-round and the laws of motion acquire centrifugal and Coriolis terms that are properties of the frame, not of any force. The quantum fictitious force is exactly as useful as its classical cousin — pick $\hat{W}$ cleverly and it cancels the awkward part of $\hat{H}$ , leaving an effective Hamiltonian that is static, or small, or zero.

Figure

The fictitious-force term — a rotating frame cancels the drive's rotation and leaves a static effective field

The master construction

Let $\hat{W}(t)$ be any differentiable family of unitaries with $\hat{W}(0)=\hat{1}$ . Define the $W$ -picture by re-draping states and operators together:

|\psi_W(t)\rangle=\hat{W}(t)\,|\psi_S(t)\rangle,\qquad \hat{A}_W(t)=\hat{W}(t)\,\hat{A}_S\,\hat{W}^\dagger(t) \qquad (1)

where: the pairing is the non-negotiable part — transform states and operators, with $\hat{W}$ and $\hat{W}^\dagger$ placed so that sandwiches collapse. (The unitary theory underwriting all of this — norm preservation, spectra under conjugation, generator doctrine — is Unitary Operators and Transformations' and is consumed here by citation.) The state's new equation of motion follows from one differentiation (Derivation A) and defines the object that distinguishes pictures from mere basis changes:

i\hbar\,\frac{d}{dt}|\psi_W\rangle=\hat{H}_W(t)\,|\psi_W\rangle,\qquad \hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\,\frac{d\hat{W}}{dt}\hat{W}^\dagger \qquad (2)

where: the first term is ordinary conjugation — how any observable transforms — and the second, $i\hbar\dot{\hat{W}}\hat{W}^\dagger$ , is the frame term: anti-hermitian-looking but in fact hermitian (differentiate $\hat{W}\hat{W}^\dagger=\hat{1}$ to see $\dot{\hat{W}}\hat{W}^\dagger$ is anti-hermitian, so $i\hbar$ times it is hermitian — $\hat{H}_W$ is a legitimate Hamiltonian). This term is the quantum fictitious force: present whenever the frame moves, absent for static $\hat{W}$ , and the lever by which pictures do useful work. The three famous pictures are three settings of the dial, each now a theorem (all three machine-checked symbolically in the verifier):

$\hat{W}(t)$	$\hat{H}_W$	picture
$\hat{1}$	$\hat{H}$	Schrödinger — states carry everything
$e^{i\hat{H}_0t/\hbar}$	$\hat{W}\hat{V}\hat{W}^\dagger=\hat{V}_I$	interaction — the frame term eats $\hat{H}_0$
$\hat{U}^\dagger(t)$	$0$	Heisenberg — the frame term eats everything; states freeze

The Heisenberg row deserves a second look. With $\hat{W}=\hat{U}^\dagger=e^{i\hat{H}t/\hbar}$ (time-independent $\hat{H}$ ), the frame term is $i\hbar\cdot\frac{i\hat{H}}{\hbar}=-\hat{H}$ , while conjugation leaves $\hat{H}$ alone; the two cancel exactly, $\hat{H}_W=0$ , and the frozen states of The Heisenberg Picture are no longer a definition but a consequence. Likewise the interaction row derives the previous article's state law: conjugation gives $\hat{H}_0+\hat{V}_I$ -in-the-frame while the frame term subtracts the transported $\hat{H}_0$ , leaving $\hat{V}_I$ alone.

Derivation AThe effective Hamiltonian of an arbitrary picture

We derive equation (2): the equation of motion obeyed by $|\psi_W\rangle=\hat{W}|\psi_S\rangle$ for arbitrary unitary $\hat{W}(t)$ . The strategy is the chapter's standard one derivative — the product rule plus the Schrödinger equation — but with no cancellation engineered in advance, so that the general structure (conjugated Hamiltonian plus frame term) appears before any special choice is made. Watch two things: where the insertion of $\hat{W}^\dagger\hat{W}=\hat{1}$ happens, and why the resulting extra term is hermitian, which is what entitles it to sit inside a Hamiltonian.

Step 1. Differentiate $|\psi_W\rangle=\hat{W}|\psi_S\rangle$ and multiply by $i\hbar$ :

i\hbar\,\frac{d}{dt}|\psi_W\rangle=i\hbar\,\dot{\hat{W}}|\psi_S\rangle+\hat{W}\,\big(i\hbar\,\partial_t|\psi_S\rangle\big)

Step 2. Insert the Schrödinger equation $i\hbar\,\partial_t|\psi_S\rangle=\hat{H}|\psi_S\rangle$ and re-express $|\psi_S\rangle=\hat{W}^\dagger|\psi_W\rangle$ in both terms:

i\hbar\,\frac{d}{dt}|\psi_W\rangle=i\hbar\,\dot{\hat{W}}\hat{W}^\dagger\,|\psi_W\rangle+\hat{W}\hat{H}\hat{W}^\dagger\,|\psi_W\rangle

Why both terms close on $|\psi_W\rangle$ : the picture map is invertible at each instant, so the old state is always recoverable from the new one — a picture never loses information.

Step 3. Read off the effective Hamiltonian:

\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\,\dot{\hat{W}}\hat{W}^\dagger

Step 4. Hermiticity of the frame term: differentiate $\hat{W}\hat{W}^\dagger=\hat{1}$ to get $\dot{\hat{W}}\hat{W}^\dagger+\hat{W}\dot{\hat{W}}^\dagger=0$ , i.e. $(\dot{\hat{W}}\hat{W}^\dagger)^\dagger=-\dot{\hat{W}}\hat{W}^\dagger$ . Anti-hermitian times $i\hbar$ is hermitian; $\hat{H}_W$ generates legitimate unitary evolution. Why this is the complete classification of pictures: any two ways of evolving the same physics differ by exactly such a $\hat{W}(t)$ — pictures form a group (compose the $\hat{W}$ 's), with Schrödinger at the identity and equation (2) as the transformation law of the "gauge field" $\hat{H}$ . *Verified: all three canonical rows of the table machine-checked symbolically on a 2×2 split Hamiltonian ( $\hat{W}=\hat{1}\Rightarrow\hat{H}$ ; $\hat{W}=e^{i\hat{H}t/\hbar}\Rightarrow0$ exactly; $\hat{W}=e^{i\hat{H}_0t/\hbar}\Rightarrow\hat{V}_I$ ), and the rotating-frame instance of the worked example verified symbolically and numerically (article verifier).*

The invariance theorem

The construction would be idle if different settings of the dial predicted different numbers. They cannot — and the proof is short enough to run in full.

Go deeper · Derivation B · 1 minEvery measurable number is picture-invariant

Derivation BEvery measurable number is picture-invariant

We prove that matrix elements, spectra, and Born probabilities take identical values in every picture, for every unitary $\hat{W}(t)$ . The strategy is cancellation: each measurable is a sandwich in which $\hat{W}^\dagger\hat{W}=\hat{1}$ collapses, so invariance is structural, not accidental. The one subtle case — probabilities of measurement outcomes when base kets themselves move (the Heisenberg bookkeeping The Heisenberg Picture tabulated) — is handled by transforming the eigenkets consistently and watching the same cancellation appear.

Step 1. Matrix elements. For any two states and any observable, using (1):

\langle\varphi_W|\hat{A}_W|\psi_W\rangle=\langle\varphi_S|\hat{W}^\dagger\,\big(\hat{W}\hat{A}_S\hat{W}^\dagger\big)\,\hat{W}|\psi_S\rangle=\langle\varphi_S|\hat{A}_S|\psi_S\rangle

Why this single line carries most of the theorem: expectation values ( $\varphi=\psi$ ), overlaps ( $\hat{A}=\hat{1}$ ), and transition amplitudes are all instances.

Step 2. Spectra. $\hat{A}_W=\hat{W}\hat{A}_S\hat{W}^\dagger$ is a unitary conjugation, and conjugation preserves eigenvalues: if $\hat{A}_S|a\rangle=a|a\rangle$ then $\hat{A}_W(\hat{W}|a\rangle)=a(\hat{W}|a\rangle)$ . The measured values of any observable — the spectrum — are picture-independent (the conjugation card of Unitary Operators and Transformations, recalled).

Step 3. Born probabilities with moving base kets. In the $W$ -picture the eigenkets of $\hat{A}_W$ are $|a\rangle_W=\hat{W}|a\rangle$ , so the outcome probability is

P_W(a,t)=\big|{}_W\langle a|\psi_W(t)\rangle\big|^2=\big|\langle a|\hat{W}^\dagger\hat{W}|\psi_S(t)\rangle\big|^2=\big|\langle a|\psi_S(t)\rangle\big|^2=P_S(a,t)

Why the Heisenberg case is the instructive extreme: there $\hat{W}=\hat{U}^\dagger$ , the base kets move backward as $|a,t\rangle_H=\hat{U}^\dagger|a\rangle$ while the state stands still, and $P(a,t)=|\langle a,t|\psi_H\rangle|^2$ — the amplitude reassembles from the other side, same number. The eigenvalue attached to the moving base ket is also unchanged, so the outcome labels themselves are frame-proof. *Verified: matrix-element and spectrum invariance machine-checked under an arbitrary random unitary family on 3×3 systems ( $10^{-12}$ ); Born probabilities with moving base kets checked in the Heisenberg extreme, including the eigenvalue-preservation of the moving base kets ( $10^{-11}$ ) — article verifier.*

What survives every re-draping is the short list worth memorizing: spectra (what values can occur), probabilities and expectation values (how often they occur), overlaps and transition amplitudes (interference). What does not survive is everything else — whether the ket moves, whether operators are static, what the effective Hamiltonian looks like, even whether energy eigenstates "rotate their phases". Those are coordinates. Physics is the invariants; the rest is stagecraft.

Limiting cases

case	result	meaning
$\hat{W}=\hat{1}$	$\hat{H}_W=\hat{H}$	Schrödinger picture — the dial's origin
$\hat{W}=\hat{U}^\dagger$	$\hat{H}_W=0$	Heisenberg picture derived; frame term cancels all of $\hat{H}$
$\hat{W}=e^{i\hat{H}_0t/\hbar}$	$\hat{H}_W=\hat{V}_I$	interaction picture derived; frame term eats $\hat{H}_0$
$\hat{W}$ static ( $\dot{\hat{W}}=0$ )	$\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger$	mere basis change — no fictitious force, not a new picture
$\hat{W}=e^{i\omega t\hat{\sigma}_z/2}$	drive de-rotated, $\hat{H}_W$ static	the rotating frame — worked example below

Common trapA picture change is not a physical transformation

Applying $\hat{W}(t)$ to states alone — without conjugating the operators — is a physical act (a rotation, a boost, a kick) that changes predictions. A picture change applies $\hat{W}$ to states and $\hat{W}\cdots\hat{W}^\dagger$ to every observable simultaneously, and changes nothing measurable. Confusing the two produces classic errors: treating the interaction picture's frozen operators as "the drive switched off", reading the Heisenberg picture's frozen states as "nothing happens", or — the laboratory version — comparing a lab-frame measured phase against a rotating-frame calculated one without undoing the frame. Always ask: did I transform one side of the sandwich, or both? One side is physics; both sides is bookkeeping.

Worked examples

Example 1 — the rotating frame that runs a qubit gate. A transmon qubit has splitting $\omega_0/2\pi=5.00$ GHz and is driven by a microwave field rotating at $\omega/2\pi=4.95$ GHz with strength $\Omega/2\pi=20$ MHz: $\hat{H}(t)=\frac{\hbar\omega_0}{2}\hat{\sigma}_z+\frac{\hbar\Omega}{2}(\cos\omega t\,\hat{\sigma}_x+\sin\omega t\,\hat{\sigma}_y)$ . Find the static picture.

Step 1. Choose the frame that co-rotates with the drive: $\hat{W}=e^{i\omega t\hat{\sigma}_z/2}$ . Conjugation turns the rotating transverse pair back into a fixed $\hat{\sigma}_x$ (the rotation undoes the drive's rotation), and leaves $\hat{\sigma}_z$ alone.

Step 2. The frame term: $i\hbar\dot{\hat{W}}\hat{W}^\dagger=-\frac{\hbar\omega}{2}\hat{\sigma}_z$ — the fictitious force, pointing opposite the splitting.

Step 3. Master formula (2), verified symbolically in the verifier:

\hat{H}_W=\frac{\hbar(\omega_0-\omega)}{2}\hat{\sigma}_z+\frac{\hbar\Omega}{2}\hat{\sigma}_x=\frac{\hbar\Delta}{2}\hat{\sigma}_z+\frac{\hbar\Omega}{2}\hat{\sigma}_x

with detuning $\Delta/2\pi=50$ MHz — time-independent. A GHz-scale driven problem became a static field tilted in the $x$ – $z$ plane.

Step 4. Read the static geometry: tilt angle $\theta=\arctan(\Omega/\Delta)=\arctan(20/50)=$ 21.8° from the $z$ -axis, effective precession frequency $\sqrt{\Delta^2+\Omega^2}/2\pi=$ 53.85 MHz. Gate designers steer $(\Delta,\Omega)$ to point this tilted field where they want the Bloch vector to go; the time-dependent dynamics that this static picture generates (Rabi flopping) belongs to Time-Dependent Perturbation Theory.

Example 2 — one probability, two pictures, same number. A spin-½ starts in $|{+}x\rangle$ under $\hat{H}=\omega\hat{S}_z$ . Compute the probability of finding $S_x=+\hbar/2$ at $\omega t=\pi/3$ , both ways.

Step 1. Schrödinger route: evolve the state, $|\psi(t)\rangle=e^{-i\omega t\hat{S}_z/\hbar}|{+}x\rangle$ ; the amplitude is $\langle{+}x|\psi(t)\rangle=\cos(\omega t/2)$ (the two $\hat{S}_z$ components pick opposite half-phases $e^{\mp i\omega t/2}$ and their overlap with $|{+}x\rangle$ averages the two).

Step 2. Heisenberg route: the state stays $|{+}x\rangle$ forever; the base ket moves backward, $|{+}x,t\rangle_H=\hat{U}^\dagger|{+}x\rangle$ , and the amplitude is $\langle{+}x,t|\psi_H\rangle=\langle{+}x|\hat{U}|{+}x\rangle$ — literally the same complex number assembled from the other side.

Step 3. Evaluate: $P(S_x{=}{+})=\cos^2(\omega t/2)=\cos^2(\pi/6)=$ 3/4 — both routes, machine-checked equal to $10^{-13}$ . The lesson generalizes: any question you can phrase as a sandwich can be computed in whichever picture makes the sandwich easiest.

Origins · 1 minCarl Eckart and how this was discovered

Carl Eckart

1902–1973 · Princeton, Caltech, Scripps

The first equivalence theorem was proven by a 24-year-old American while the two theories it unified were still glaring at each other across the Atlantic. In early 1926 physics had two quantum mechanics: Heisenberg's matrices (Göttingen) and Schrödinger's waves (Zürich), with partisans trading barbs about which was the theory. Eckart, a fresh Caltech postdoc, built an operator calculus in which both appeared as representations of one structure and published "Operator calculus and the solution of the equations of quantum dynamics" — proving, independently of Schrödinger's own equivalence paper of the same spring, that the rivals were one theory in two dressings. It was this article's theorem in embryo: pictures differ, sandwiches agree. Eckart's later career kept the theme of translating between registers — the Wigner–Eckart theorem factors matrix elements into geometry times physics, and quantum chemists know the Eckart frame, the rotating molecular frame chosen to decouple vibration from rotation: a picture change in the mechanical sense. From 1942 he turned to the ocean, directing sonar research and founding modern physical oceanography at Scripps — underwater acoustics being, fittingly, one more exercise in choosing the frame in which a wave problem becomes tractable.

History

1925

— Heisenberg's matrix mechanics: dynamics carried by time-dependent matrices

1926

— Schrödinger's wave mechanics — and his equivalence paper; Eckart's operator-calculus proof arrives independently the same spring

1926

— Dirac's transformation theory frames both as coordinate choices on one state space

1927

— Dirac's radiation paper adds the intermediate picture between the two poles

1932

— von Neumann's Grundlagen makes unitary equivalence the mathematical backbone

1957

— Rabi-era rotating-frame methods make picture engineering standard laboratory practice

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §2.2 & §5.5 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 4 & §18.3 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I Ch. III & Vol. II Ch. XIII · Eckart, Phys. Rev. 28, 711 (1926) · Messiah, Quantum Mechanics, Ch. VIII (representations and pictures).

Provenance. All structural results are standard. A verifier should reproduce: (1) the master equivalence by inserting $|\psi_S(t)\rangle=U|\psi_H\rangle$ into a matrix element (Sakurai §2.2); (2) the Heisenberg EOM by differentiating $A_H=U^\dagger A_S U$ with $i\hbar\,\partial_t U=HU$ (Sakurai §2.2, Shankar Ch. 4); (3) the interaction-picture EOM via $|\psi_I\rangle=U_0^\dagger|\psi_S\rangle$ (Sakurai §5.5, Cohen-Tannoudji Complement); (4) the factorization $U=U_0 U_I$ and the time-ordered exponential (Dyson series). No numerical constants are asserted beyond the exact factor $1/(i\hbar)$ in the equations of motion. For this edition, items (1)–(3) were machine-executed (sympy/numpy): the master formula $\hat{H}_W=\hat{W}\hat{H}\hat{W}^\dagger+i\hbar\dot{\hat{W}}\hat{W}^\dagger$ on all three canonical $\hat{W}$ 's (Schrödinger/Heisenberg/interaction rows recovered exactly, symbolic); matrix-element and spectrum invariance under an arbitrary random unitary family ( $10^{-12}$ ); Born probabilities with moving base kets in the Heisenberg extreme ( $10^{-13}$ ); the rotating-frame reduction and the transmon numbers (21.8°, 53.85 MHz); the two-picture probability 3/4. Item (4) is consumed from Time-Ordered Exponential and the Dyson Series per the chapter ownership map; the Heisenberg EOM of item (2) is owned by The Heisenberg Equation of Motion (recalled); the S/H/I picture contracts by their own articles; Rabi dynamics of the rotating frame by TDEP.

Builds on: The Schrodinger Picture · The Heisenberg Picture · The Interaction (Dirac) Picture · Unitary Operators and Transformations · The Time-Evolution Operator

Leads to: Ehrenfest's Theorem · Constants of the Motion and Conservation Laws · Symmetries and Conservation in Quantum Dynamics · Time-Dependent Perturbation Theory

If all pictures are equivalent, why does anyone bother proving it?

Because the equivalence is used silently everywhere, and an unproven invariance is a landmine. Every rotating-frame gate design computes in one picture and measures in another; every QFT calculation mixes interaction-picture fields with Heisenberg-picture currents. The theorem is what entitles those mixtures to a common answer — and its proof (cancelling unitaries in sandwiches) tells you exactly which objects are safe to transport (sandwiches) and which are not (bare kets, effective Hamiltonians, phases of individual amplitudes).

Is the effective Hamiltonian H_W an observable? Its value seems to depend on the picture.

It is a generator, not an invariant. $\hat{H}_W$ 's job is to move $|\psi_W\rangle$ ; its spectrum has no picture-independent meaning ( $\hat{H}_W=0$ in the Heisenberg picture while the system's energies are unchanged!). The observable "energy" is the original $\hat{H}$ conjugated like any operator, $\hat{W}\hat{H}\hat{W}^\dagger$ , whose spectrum is invariant. The two differ by exactly the frame term — which is why rotating-frame "energies" (detunings, dressed splittings) are frame artifacts, meaningful only with the frame declared.

Why is the frame term hermitian when Ẇ W† looks like it shouldn't be?

Differentiate the unitarity constraint $\hat{W}\hat{W}^\dagger=\hat{1}$ : $\dot{\hat{W}}\hat{W}^\dagger=-(\dot{\hat{W}}\hat{W}^\dagger)^\dagger$ , so $\dot{\hat{W}}\hat{W}^\dagger$ is anti-hermitian — it lives in the Lie algebra of the unitary group, the same place $-i\hat{G}/\hbar$ lives for any hermitian generator $\hat{G}$ . Multiplying by $i\hbar$ rotates anti-hermitian into hermitian. The classical analogue is exact: a rotating frame's angular velocity enters the effective force law as a real (fictitious) force, not an imaginary one.

Do pictures form a group? What happens if I compose two picture changes?

Yes — compose the unitaries: going from Schrödinger to $W_1$ then applying $W_2$ lands at $W_2W_1$ , another lawful picture. The master formula behaves like a gauge transformation under composition: apply it twice and the frame terms add appropriately. This is used in earnest in qubit control (drive frame, then the interaction frame of what remains, then an averaging frame) — each stage strips a scale from the problem, and the master formula is the recursion step.

Does the equivalence still hold when the Hamiltonian depends on time?

Yes — nothing in Derivation A assumed a static $\hat{H}$ ; the Schrödinger equation was inserted as-is. Two caveats inherit from elsewhere: the Heisenberg row's clean cancellation $\hat{H}_W=0$ uses $\hat{W}=\hat{U}^\dagger$ with $\hat{U}$ the full (time-ordered) evolution operator — still exact, just harder to write; and closed forms for $\hat{U}$ itself need the Dyson machinery when $[\hat{H}(t),\hat{H}(t')]\neq0$ . The invariance theorem is indifferent: it only ever used unitarity.

Is choosing a picture the same as choosing a gauge?

Structurally, yes — and the analogy is precise enough to compute with. The state's "covariant derivative" $i\hbar\partial_t-\hat{H}$ transforms under $\hat{W}(t)$ exactly like a gauge-covariant derivative under a local gauge transformation, with $\hat{H}$ playing the gauge potential and the frame term the inhomogeneous piece. Electromagnetic gauge changes $\psi\to e^{iq\chi(t)/\hbar}\psi$ are literally picture changes with $\hat{W}$ a phase; the full gauge story (minimal coupling, local $\chi(x,t)$ ) belongs to the relativistic track, but the habit of asking "what is invariant?" transfers unchanged.

Which picture should I actually compute in?

Let the problem's structure choose. Wavefunction evolution, numerical PDE work, tunneling: Schrödinger. Operator algebra, conservation laws, correlation functions, field theory: Heisenberg. Driven systems with a solvable backbone: interaction/rotating frame — and engineer $\hat{W}$ until $\hat{H}_W$ is static or small. The invariance theorem is your license to switch mid-problem: compute each factor of a sandwich in whichever frame it is cheapest, then assemble — the number cannot depend on the itinerary.