The Interaction (Dirac) Picture

Two pictures now stand at opposite poles: Schrödinger, where states carry every scrap of time dependence, and Heisenberg, where operators carry it all. Real problems tend to arrive in a shape neither pole loves. An atom in a laser field, a spin under a radio pulse, a nucleus prodded by a passing projectile — each is a Hamiltonian that is mostly solved plus a piece that is not: $\hat{H}=\hat{H}_0+\hat{V}(t)$ , with $\hat{H}_0$ diagonalized long ago and $\hat{V}$ the interesting intruder. Dirac's 1927 move — the interaction picture — is a division of labor that matches this shape exactly. Hand the solved part to the operators and let $\hat{H}_0$ rotate them, exactly as in the Heisenberg picture of a free theory. Hand the unsolved part to the states: the interaction-picture ket obeys a Schrödinger equation whose only Hamiltonian is the transformed perturbation $\hat{V}_I(t)$ . Everything already understood is spent; everything left to compute is isolated. When $\hat{V}$ is weak, the state barely moves, and slow motion is precisely what perturbation theory integrates well. The mechanics runs on one exponential, $e^{i\hat{H}_0t/\hbar}$ , and produces one signature everywhere: matrix elements of $\hat{V}_I$ beat at Bohr frequencies, $\langle m|\hat{V}_I|n\rangle=V_{mn}e^{i\omega_{mn}t}$ — the phases whose resonances decide which transitions a drive can actually cause. This article derives the two equations of motion of the picture, assembles the Dyson series where it is actually used, verifies first order against an exact two-level integration, and states the covariant generalization for which Tomonaga rebuilt the whole idea. Time-dependent perturbation theory, four chapters on, is this article turned into an industry.

Key takeaways

Split $\hat{H}=\hat{H}_0+\hat{V}(t)$ : the interaction picture assigns $\hat{H}_0$ 's motion to operators and $\hat{V}$ 's to states — $|\psi_I\rangle=e^{i\hat{H}_0t/\hbar}|\psi_S\rangle$ , $\hat{A}_I=e^{i\hat{H}_0t/\hbar}\hat{A}_Se^{-i\hat{H}_0t/\hbar}$ .
States obey $i\hbar\,\partial_t|\psi_I\rangle=\hat{V}_I(t)|\psi_I\rangle$ with $\hat{V}_I=e^{i\hat{H}_0t/\hbar}\hat{V}e^{-i\hat{H}_0t/\hbar}$ ; operators obey $\dot{\hat{A}}_I=\frac{i}{\hbar}[\hat{H}_0,\hat{A}_I]$ .
Matrix elements of the moving perturbation beat at Bohr frequencies: $\langle m|\hat{V}_I|n\rangle=V_{mn}\,e^{i\omega_{mn}t}$ , $\omega_{mn}=(E_m-E_n)/\hbar$ .
$\hat{V}\to0$ freezes the states (Heisenberg limit for the free theory); $\hat{H}_0\to0$ returns the Schrödinger picture — the interaction picture interpolates.
$\hat{U}_I$ solves an integral equation whose iteration is the Dyson series; first order gives $c_f^{(1)}(t)=-\frac{i}{\hbar}\int_0^t V_{fi}(t')e^{i\omega_{fi}t'}dt'$ — the seed of all of time-dependent perturbation theory.
The full evolution factorizes as $\hat{U}(t)=e^{-i\hat{H}_0t/\hbar}\,\hat{U}_I(t)$ : free rotation times slow drift.
Key idea: spend the solved physics on the operators, isolate the unsolved physics in the states — the perturbation, dressed with Bohr-frequency phases, is the only thing left moving the ket.

The core insight

Split $\hat{H}=\hat{H}_0+\hat{V}$ and let each part run its own picture: operators rotate under $\hat{H}_0$ alone, states drift under the dressed perturbation $\hat{V}_I(t)=e^{i\hat{H}_0t/\hbar}\hat{V}e^{-i\hat{H}_0t/\hbar}$ alone — the natural frame for perturbation theory.

Where this lives today

Every field-theory calculation begins with this frame: the S-matrix of particle physics is $\hat{U}_I(+\infty,-\infty)$ , free fields playing $\hat{H}_0$ and the coupling playing $\hat{V}$ , with Feynman diagrams as the pictures of Dyson-series terms in Dirac's frame. A caveat with its own literature (Haag's theorem) says the frame is subtler in infinite volume — one reason rigorous QFT still studies exactly this construction. In the laboratory the picture wears its NMR name, "the rotating frame": every pulse sequence in a hospital MRI scanner, every gate on a superconducting or trapped-ion qubit is designed in interaction coordinates, where a GHz-scale carrier disappears and only MHz-scale control physics remains. When a hardware paper writes "in the frame rotating at the drive frequency, the effective Hamiltonian is…", it is executing equation (1) — and cashing exactly the frequency-separation dividend of Example 1.

The physical picture

Think of a spinning carousel with a child walking slowly across it. Watched from the ground (Schrödinger), the child traces a dizzy spiral — fast rotation and slow walk hopelessly entangled. Climb aboard the carousel (interaction picture) and the rotation vanishes from view: what remains is the walk, slow and legible. The carousel is $\hat{H}_0$ — known, fast, exactly solvable; the walk is $\hat{V}$ — unknown, slow, the thing you actually want to study. The frame change $e^{i\hat{H}_0t/\hbar}$ is the climbing aboard.

Figure

Divide the labor — H₀ spins the operators, the dressed perturbation walks the states

The frame purchase is largest when the two rates are far apart, and in practice they usually are. An optical transition beats at $\omega_0/2\pi\sim500$ THz while a laboratory drive nudges populations over microseconds — nine orders of magnitude of separation. Integrating the raw Schrödinger equation numerically means resolving the fast phase for a billion cycles to see one slow transfer; in the interaction picture the fast phase is analytic (it lives inside $\hat{V}_I$ 's Bohr factors) and only the slow transfer is left to compute. That is not merely computational hygiene — it is why resonance is visible at a glance: a drive component oscillating at $\omega$ makes $\langle f|\hat{V}_I|i\rangle\sim e^{i(\omega_{fi}-\omega)t}$ nearly stationary when $\omega\approx\omega_{fi}$ , and stationary phases are the ones whose integrals grow.

Figure

Bohr-frequency phases — off resonance the amplitude stirs in circles, on resonance it walks

Definitions and the two laws

Fix the split $\hat{H}=\hat{H}_0+\hat{V}(t)$ , with $\hat{H}_0$ time-independent and solved: eigenkets $|n\rangle$ , energies $E_n$ , free evolution $\hat{U}_0(t)=e^{-i\hat{H}_0t/\hbar}$ (The Time-Evolution Operator's exponential). The interaction picture unwinds exactly that free motion from the Schrödinger state:

|\psi_I(t)\rangle=e^{i\hat{H}_0t/\hbar}\,|\psi_S(t)\rangle,\qquad \hat{A}_I(t)=e^{i\hat{H}_0t/\hbar}\,\hat{A}_S\,e^{-i\hat{H}_0t/\hbar} \qquad (1)

where: the operator dressing uses the same unitary for every observable — including the perturbation itself, whose dressed form $\hat{V}_I(t)=e^{i\hat{H}_0t/\hbar}\hat{V}(t)e^{-i\hat{H}_0t/\hbar}$ will run the show. At $t=0$ all three pictures coincide; the sign conventions (a $+i$ for kets, conjugation for operators) are fixed by demanding that matrix elements — the physics — stay put, the invariance Equivalence of the Dynamical Pictures proves next. Differentiating the two definitions gives the picture's two laws (Derivation A):

i\hbar\,\frac{d}{dt}|\psi_I(t)\rangle=\hat{V}_I(t)\,|\psi_I(t)\rangle \qquad (2)

where: the free Hamiltonian has cancelled out of the state's equation — the ket feels only the dressed perturbation. If $\hat{V}$ is weak, $|\psi_I\rangle$ is nearly frozen; if $\hat{V}=0$ it is exactly frozen, and the picture degenerates into the Heisenberg picture of the free theory.

\frac{d\hat{A}_I}{dt}=\frac{i}{\hbar}\,[\hat{H}_0,\hat{A}_I]+\Big(\frac{\partial\hat{A}}{\partial t}\Big)_I \qquad (3)

where: operators obey a Heisenberg equation (The Heisenberg Equation of Motion, previous article) whose generator is $\hat{H}_0$ alone — the solved part. Equations (2) and (3) together are the division of labor: nothing about $\hat{H}$ is used twice, and nothing is dropped, since the full evolution factorizes as $\hat{U}(t)=\hat{U}_0(t)\hat{U}_I(t)$ . The observable consequence of the dressing is one phase. Take matrix elements of $\hat{V}_I$ in the free eigenbasis:

\langle m|\hat{V}_I(t)|n\rangle=V_{mn}(t)\,e^{i\omega_{mn}t},\qquad \omega_{mn}=\frac{E_m-E_n}{\hbar} \qquad (4)

where: $V_{mn}(t)=\langle m|\hat{V}(t)|n\rangle$ is the bare matrix element and the exponential is the Bohr-frequency beat — the same frequencies Spectral Decomposition of Time Evolution showed are the only ones observables can oscillate at. Equation (4) is the workhorse of spectroscopy-adjacent quantum mechanics: a drive transfers amplitude efficiently only where its own oscillation cancels this phase.

Derivation AThe two equations of motion of the interaction picture

We derive equations (2) and (3) from the definitions (1). The state law is the heart: the strategy is to differentiate $|\psi_I\rangle=e^{i\hat{H}_0t/\hbar}|\psi_S\rangle$ and watch the $\hat{H}_0$ generated by the frame exactly cancel the $\hat{H}_0$ inside the Schrödinger equation, leaving only the perturbation — dressed by the frame on its way out. The point to watch is where $e^{i\hat{H}_0t/\hbar}$ lands relative to $\hat{V}$ : it does not commute with $\hat{V}$ in general, and the dressing $\hat{V}_I=e^{i\hat{H}_0t/\hbar}\hat{V}e^{-i\hat{H}_0t/\hbar}$ is manufactured by an insertion of unity, the same move that closed the Heisenberg derivation.

Step 1. Differentiate the state definition; two terms, one from the frame, one from the moving Schrödinger ket:

i\hbar\,\frac{d}{dt}|\psi_I\rangle=-\hat{H}_0\,e^{i\hat{H}_0t/\hbar}|\psi_S\rangle+e^{i\hat{H}_0t/\hbar}\,\big(i\hbar\,\partial_t|\psi_S\rangle\big)

Why the first sign: $i\hbar\frac{d}{dt}e^{i\hat{H}_0t/\hbar}=i\hbar\cdot\frac{i\hat{H}_0}{\hbar}e^{i\hat{H}_0t/\hbar}=-\hat{H}_0e^{i\hat{H}_0t/\hbar}$ — the frame's derivative arrives with a minus, primed to cancel.

Step 2. Insert the Schrödinger equation $i\hbar\,\partial_t|\psi_S\rangle=(\hat{H}_0+\hat{V})|\psi_S\rangle$ :

i\hbar\,\frac{d}{dt}|\psi_I\rangle=-\hat{H}_0|\psi_I\rangle+e^{i\hat{H}_0t/\hbar}(\hat{H}_0+\hat{V})|\psi_S\rangle

Step 3. The $\hat{H}_0$ pieces cancel ( $e^{i\hat{H}_0t/\hbar}$ commutes with $\hat{H}_0$ ), and unity $e^{-i\hat{H}_0t/\hbar}e^{i\hat{H}_0t/\hbar}$ inserted after $\hat{V}$ dresses both factors:

i\hbar\,\frac{d}{dt}|\psi_I\rangle=e^{i\hat{H}_0t/\hbar}\,\hat{V}\,e^{-i\hat{H}_0t/\hbar}\;e^{i\hat{H}_0t/\hbar}|\psi_S\rangle=\hat{V}_I(t)\,|\psi_I\rangle

Why the cancellation is the whole point: the frame was chosen precisely to generate $-\hat{H}_0$ ; any other choice would leave a residue. What survives is the one term the frame cannot absorb — the unsolved physics.

Step 4. For the operator law, differentiate $\hat{A}_I=e^{i\hat{H}_0t/\hbar}\hat{A}_Se^{-i\hat{H}_0t/\hbar}$ exactly as in the Heisenberg derivation with $\hat{H}\to\hat{H}_0$ (the frame unitary solves $i\hbar\dot{\hat{U}}_0=\hat{H}_0\hat{U}_0$ ):

\frac{d\hat{A}_I}{dt}=\frac{i}{\hbar}[\hat{H}_0,\hat{A}_I]+\Big(\frac{\partial\hat{A}}{\partial t}\Big)_I

Verified: the reconstruction $\hat{U}(t)=e^{-i\hat{H}_0t/\hbar}\hat{U}_I(t)$ machine-checked against a high-accuracy direct integration of the full two-level Schrödinger equation at three times ( $10^{-8}$ ); the operator law checked on a time grid ( $10^{-3}$ central differences); the Bohr phase of equation (4) symbolic — all in the article's verifier.

Where Dyson's series is actually used

Equation (2) has no closed solution in general — $\hat{V}_I(t)$ at different times need not commute. Its integral form,

\hat{U}_I(t)=\hat{1}-\frac{i}{\hbar}\int_0^t \hat{V}_I(t')\,\hat{U}_I(t')\,dt' \qquad (5)

where: $\hat{U}_I$ is the interaction-picture evolution operator ( $|\psi_I(t)\rangle=\hat{U}_I(t)|\psi_I(0)\rangle$ ), invites iteration: substitute the left side into the right, repeatedly. The result is the Dyson series — and the construction, the time-ordering symbol, and the $1/n!$ combinatorics were all built two articles ago in Time-Ordered Exponential and the Dyson Series; we consume the machinery rather than rebuild it. What the interaction picture adds is the reason the series converges usefully: each factor is a dressed perturbation, small by hypothesis, so successive terms are genuinely higher order in the weak coupling. Keeping only the first iteration and sandwiching between free eigenstates gives the celebrated first-order transition amplitude:

c^{(1)}_f(t)=-\frac{i}{\hbar}\int_0^t \langle f|\hat{V}(t')|i\rangle\,e^{i\omega_{fi}t'}\,dt' \qquad (6)

where: the amplitude is the Fourier component of the drive at the Bohr frequency — resonance made quantitative. Equation (6) is the bridge this chapter hands to Time-Dependent Perturbation Theory: transition probabilities, the $\sin^2$ line shape, Fermi's golden rule and Rabi's full resummation all begin here, and none are owned here.

Go deeper · Derivation B · 1 minFirst order from the integral equation

Derivation BFirst order from the integral equation

We derive the transition amplitude (6) by iterating the integral equation (5) exactly once and projecting onto free eigenstates. The strategy is the standard perturbative seed: replace the unknown $\hat{U}_I$ under the integral by its zeroth approximation $\hat{1}$ , which is justified precisely when $\hat{V}_I$ is weak and $t$ not too long — the correction to this replacement is second order by construction. Watch how the Bohr phase of equation (4) enters: it is not an extra assumption but the automatic fingerprint of the dressing, and it is the entire resonance physics of the result.

Step 1. Iterate once — substitute $\hat{U}_I(t')\approx\hat{1}$ under the integral of (5):

\hat{U}_I(t)\approx\hat{1}-\frac{i}{\hbar}\int_0^t \hat{V}_I(t')\,dt'

Why the error is second order: the exact $\hat{U}_I(t')$ differs from $\hat{1}$ by a term itself of order $\hat{V}_I$ ; inserted under the integral it produces the double- $\hat{V}$ contribution — the Dyson series' next term, owned by the series article.

Step 2. Take the transition matrix element between distinct free eigenstates $|i\rangle\to|f\rangle$ ( $\langle f|i\rangle=0$ kills the identity):

c_f(t)=\langle f|\hat{U}_I(t)|i\rangle\approx-\frac{i}{\hbar}\int_0^t \langle f|\hat{V}_I(t')|i\rangle\,dt'

Step 3. Undress the matrix element with equation (4), $\langle f|\hat{V}_I(t')|i\rangle=V_{fi}(t')e^{i\omega_{fi}t'}$ :

c_f^{(1)}(t)=-\frac{i}{\hbar}\int_0^t V_{fi}(t')\,e^{i\omega_{fi}t'}\,dt'

Why this reads as a Fourier statement: the amplitude is the drive's matrix element analysed at exactly the Bohr frequency of the transition — a drive moves population efficiently only where its spectrum overlaps $\omega_{fi}$ , the stationary-phase reading of the second figure. *Verified: for a weak near-resonant harmonic drive on a two-level system, the quadrature-evaluated Step-3 integral matches a high-accuracy direct integration of the full Schrödinger equation to 0.2% at $|c_f|\approx0.08$ (article verifier) — and the agreement degrades as the drive strengthens, the honest boundary of first order.*

The interpolation table

The picture triad closes into one family. The Schrodinger Picture and The Heisenberg Picture built the poles and their decision table; the interaction picture is the tunable middle:

	states carry	operators carry	natural problems
Schrödinger	all of $\hat{H}$	nothing (static)	wavefunction dynamics, numerics
interaction	$\hat{V}_I$ only	$\hat{H}_0$ only	driven systems, perturbation theory, QFT
Heisenberg	nothing (frozen)	all of $\hat{H}$	operator algebra, conservation, field theory

Slide $\hat{V}\to0$ and the middle row collapses onto the bottom (states freeze); slide $\hat{H}_0\to0$ and it collapses onto the top ( $\hat{V}_I\to\hat{V}$ , the frame stops turning). Both limits are machine-checked trivially in the verifier — and both are worth internalizing, because mis-assigned time dependence is the most common practical error in this subject.

Limiting cases

case	result	meaning
$\hat{V}=0$	$\|\psi_I\rangle$ frozen; $\hat{U}_I=\hat{1}$	Heisenberg picture of the free theory
$\hat{H}_0=0$	$\hat{V}_I=\hat{V}$ ; picture = Schrödinger	the frame stops turning
$[\hat{H}_0,\hat{V}]=0$	$\hat{V}_I(t)=\hat{V}(t)$ , phases gone	no Bohr beats — $\hat{V}$ cannot drive transitions between $\hat{H}_0$ levels it commutes with
weak $\hat{V}$ , short $t$	first order (6) accurate	verifier: 0.2% at $\|c_f\|\approx0.08$
$\omega\to\omega_{fi}$	stationary Bohr phase	resonant growth of $c_f$ — spectroscopy's selection principle

Common trapThe interaction picture does not neglect the coupling

"Operators evolve freely" tempts a fatal misreading — that the picture is itself an approximation which drops $\hat{V}$ 's effect on observables. Nothing is dropped anywhere: the factorization $\hat{U}=\hat{U}_0\hat{U}_I$ is exact, and the perturbation's entire influence is present, relocated into the state's drift under $\hat{V}_I$ . The approximation, when one is made, is a separate act: truncating the Dyson series for $\hat{U}_I$ . Conflating the exact frame change with the perturbative truncation causes real errors — e.g. believing interaction-picture operators "miss" level shifts, or double-counting $\hat{H}_0$ by evolving states with it too. Frame exact, truncation optional: keep the two decisions apart.

Worked examples

Example 1 — proton NMR: the picture earns its keep. A proton sits in $B_0=1$ T ( $\hat{H}_0=-\gamma B_0\hat{S}_z$ , $\gamma=2.675\times10^{8}\ \mathrm{rad\,s^{-1}T^{-1}}$ ), driven by a weak transverse field $B_1=10\ \mathrm{\mu T}$ oscillating on resonance. How much population moves in a $100\ \mathrm{\mu s}$ pulse?

Step 1. The free rate: $f_0=\gamma B_0/2\pi=42.58$ MHz — the Larmor precession that the interaction frame absorbs. On the carousel, spin operators rotate at $f_0$ ; the state feels only the drive.

Step 2. The dressed drive: the linear oscillation splits into co- and counter-rotating halves; near resonance the co-rotating half has a stationary Bohr phase and dominates (the counter-rotating half curls up — Example figure 2's logic). Its strength is $\omega_1/2=\gamma B_1/2=1338\ \mathrm{rad/s}$ .

Step 3. First order, equation (6) with stationary phase: $|c_f^{(1)}|=\frac{\gamma B_1}{2}\,t=(1338)(10^{-4})=0.134$ .

Step 4. Transition probability $P_f=|c_f|^2=$ 1.8% — and the exact two-level integration (no rotating-wave shortcut) gives 1.78%, confirming first order to 3% of itself. The frequency hierarchy is the lesson: the state's motion (kHz scale) sits five decades below the frame's (42.58 MHz), exactly the separation the picture is built to exploit. The full Rabi resummation belongs to Time-Dependent Perturbation Theory.

Example 2 — reading Bohr phases as clocks. A two-level system has splitting $E_f-E_i=1$ eV. What are the timescales of its interaction-picture bookkeeping under a drive detuned by 1%?

Step 1. The Bohr phase $e^{i\omega_{fi}t}$ turns once every $T_0=h/(1\ \mathrm{eV})=4.14$ fs — the fast clock that lives analytically inside $\hat{V}_I$ , never integrated numerically.

Step 2. Against a drive at $\hbar\omega=0.99$ eV, the surviving phase is $e^{i(\omega_{fi}-\omega)t}$ , turning once every $T_{\rm beat}=h/(0.01\ \mathrm{eV})=$ 414 fs — a hundredfold slower.

Step 3. The interaction-picture state drifts on the beat clock, not the Bohr clock: a numerical solver may take steps a hundred times longer, and the resonance condition is read off as "beat clock → infinity". That factor is the practical content of the picture: the closer to resonance, the greater the advantage of working in Dirac's frame.

Origins · 1 minSin-Itiro Tomonaga and how this was discovered

Sin-Itiro Tomonaga

1906–1979 · Kyoto, Leipzig, Tokyo

The interaction picture's deepest reading — not a convenience but a statement about what "the state at a time" can mean — came from a physicist working in near-total isolation. Tomonaga, son of a Kyoto philosophy professor and schoolmate of Yukawa, had learned quantum theory in Heisenberg's Leipzig; back in wartime Tokyo, cut off from Western journals, he confronted a flaw in Dirac's picture as relativity sees it: $|\psi_I(t)\rangle$ assigns one ket to an entire constant-time slice of the universe, a manifestly frame-dependent object. His 1943 "super-many-time" theory (published in Japanese in 1943, in English in 1946) replaced the single $t$ with a whole spacelike surface $\sigma$ , one local clock per point, and rewrote equation (2) as the covariant Tomonaga–Schwinger equation $i\hbar c\,\delta|\psi[\sigma]\rangle/\delta\sigma(x)=\mathcal{H}_I(x)|\psi[\sigma]\rangle$ — the interaction picture made relativistic, with the dressed density $\mathcal{H}_I$ doing precisely $\hat{V}_I$ 's job at each spacetime point. When the Shelter Island results (the Lamb shift) reached Japan through a magazine article, his group computed with this machinery in months. The 1965 Nobel Prize, shared with Schwinger and Feynman, honored a frame change conceived while Tokyo burned.

History

1926

— Schrödinger and Heisenberg pictures crystallize; the poles of the time-dependence spectrum stand

1927

— Dirac's radiation-theory paper introduces the intermediate picture and first-order transition amplitudes

1943

— Tomonaga's super-many-time theory: the interaction picture made covariant, one clock per spacetime point

1946

— Tomonaga–Schwinger equation reaches the West; interaction-picture QED begins in earnest

1949

— Dyson unifies the covariant formalisms; the I-picture series becomes QED's computational engine

1957

— NMR and coherent spectroscopy adopt the rotating frame as laboratory practice

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §5.5–5.6 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. II, Ch. XIII · Shankar, Principles of Quantum Mechanics, 2nd ed., §18.3 · Dirac, Proc. R. Soc. A 114, 243 (1927) · Tomonaga, Prog. Theor. Phys. 1, 27 (1946) · Schweber, QED and the Men Who Made It, Ch. 6 (Tomonaga).

Provenance. Definitions and equations follow Sakurai & Napolitano §5.5–5.6, Cohen-Tannoudji Vol. II Ch. XIII, and Shankar §18.3. A verifier should recompute: (1) the cancellation of $H_0$ terms in the state-equation derivation yielding $i\hbar\partial_t|\psi_I\rangle=V_I|\psi_I\rangle$ ; (2) the Bohr-frequency phase $e^{i\omega_{mn}t}$ in the matrix element $\langle m|V_I|n\rangle$ ; (3) the Dyson series nested-integral structure; (4) the long-time limit $4\sin^2(\omega t/2)/(\omega^2 t)\to 2\pi\delta(\omega)$ giving the golden-rule rate $\Gamma=\tfrac{2\pi}{\hbar}|V_{fi}|^2\rho(E_f)$ . No nonstandard numerical constants are used. For this edition, items (1)–(3) were machine-executed (sympy/numpy/scipy): the reconstruction $U=U_0U_I$ against exact integration ( $10^{-8}$ ), the Bohr phase symbolically, the operator law on a time grid, first order vs exact for a weak near-resonant drive (0.2%), and the NMR numbers (42.58 MHz, 2675 rad/s at 10 μT, $P_f=1.8\%$ at 100 μs vs exact 1.78%) from CODATA $\gamma_p$ . Item (4) — the golden-rule limit — is deliberately NOT executed here: it is owned by Time-Dependent Perturbation Theory, where the line-shape analysis lives. Per the chapter ownership map: the Dyson-series construction and time-ordering combinatorics belong to Time-Ordered Exponential and the Dyson Series (consumed with citation); the picture-equivalence proof to Equivalence of the Dynamical Pictures (next serial); Rabi resummation and golden rule to TDEP; the S/H picture contracts to their own articles (recalled).

Builds on: The Time-Evolution Operator · The Schrodinger Picture · The Heisenberg Picture · Time-Ordered Exponential and the Dyson Series · The Heisenberg Equation of Motion

Leads to: Equivalence of the Dynamical Pictures · Time-Dependent Perturbation Theory · Fermi's Golden Rule · The Quantum Propagator

Is the interaction picture an approximation?

No — it is an exact change of bookkeeping, valid for arbitrarily strong $\hat{V}$ . The factorization $\hat{U}=\hat{U}_0\hat{U}_I$ holds identically; nothing is neglected. Approximation enters only if you truncate the Dyson series for $\hat{U}_I$ , which is a separate decision the picture merely makes convenient. For strong coupling the picture is still exact but less useful, because $|\psi_I\rangle$ no longer moves slowly and the series no longer converges quickly.

How do I choose the split H = H₀ + V when it isn't handed to me?

Three criteria, in order: $\hat{H}_0$ must be solvable (you need its eigendata to build $e^{i\hat{H}_0t/\hbar}$ and the Bohr phases); $\hat{V}$ should be small in the matrix-element sense $|V_{mn}|\ll\hbar|\omega_{mn}|$ for the levels you care about; and the split should match the question — the same atom-plus-laser problem uses atom as $\hat{H}_0$ for absorption rates but atom-plus-field ("dressed states") as $\hat{H}_0$ for strong driving. A bad split costs convergence, never correctness.

Why does V_I carry the Bohr phases rather than the drive's own frequency?

Because the dressing conjugates by the system's free evolution, which knows only $E_m-E_n$ . The drive's frequency lives inside the bare $V_{mn}(t)$ factor. The two meet in the product $V_{mn}(t)e^{i\omega_{mn}t}$ : for a harmonic drive $\cos\omega t$ this contains $e^{i(\omega_{mn}\pm\omega)t}$ , and the near-stationary combination is the resonant, population-moving one. The picture separates what the system offers (Bohr frequencies) from what you supply (drive spectrum) and shows transfer where they match.

What happened to the counter-rotating term everyone drops in NMR?

It is the $e^{i(\omega_{fi}+\omega)t}$ half of the dressed harmonic drive — a phasor turning at nearly twice the Larmor frequency, whose integral stays bounded and tiny (the curling circle of the second figure). Dropping it is the rotating-wave approximation, a further step beyond the exact picture, good when $\omega_1\ll\omega_0$ (Example 1 has five orders of margin). Its small residue, the Bloch–Siegert shift, is measurable in precision experiments — a nice reminder that the picture is exact and the RWA is not.

Are interaction-picture operators observable? Whose expectation values do they give?

They are the same observables in a different dress: $\langle\psi_I|\hat{A}_I|\psi_I\rangle=\langle\psi_S|\hat{A}_S|\psi_S\rangle$ identically, because the frame unitary cancels between bra, operator and ket. What you must not do is mix registers — $\langle\psi_I|\hat{A}_S|\psi_I\rangle$ is not anything physical. Every prediction is picture-invariant; the proof of that statement in full generality is exactly the next article's business.

Does the interaction picture survive in quantum field theory?

It is the working language of perturbative QFT — free fields as $\hat{H}_0$ , interactions as $\hat{V}$ , S-matrix as $\hat{U}_I$ across all time, Feynman rules as Dyson-series bookkeeping. Two honest footnotes: relativity requires Tomonaga's surface generalization (one clock per spacetime point) for manifest covariance, and Haag's theorem shows the picture's naive Hilbert-space assumptions fail in interacting infinite-volume QFT — perturbation theory works anyway, for reasons renormalization theory makes precise. Both subtleties live in the RQM/QFT track.

When should I abandon the interaction picture for a numerical solver in the Schrödinger picture?

When there is no frequency hierarchy to exploit: strong driving ( $\omega_1\sim\omega_0$ ), fast chirps that sweep through many resonances, or dense spectra where dressing produces no simplification. A raw solver then integrates the full $\hat{H}(t)$ with steps resolving the fastest scale present — costly but assumption-free. The practical crossover in most driven-qubit work sits around $\omega_1/\omega_0\sim0.1$ , where RWA errors reach the percent level and "exact numerics in the lab frame" becomes the standard check on frame-based analytics.