The Time-Evolution Operator

The formalism chapter built the stage: states as vectors, observables as Hermitian operators, measurement as projection. This chapter sets the stage in motion — and opens with a change of strategy. Instead of solving the Schrödinger equation state by state, quantum dynamics packages all motion into one object: the time-evolution operator $\hat U(t, t_0)$ , the unitary map taking any state at $t_0$ to its value at $t$ . Solve for $\hat U$ once and every initial condition is solved forever. The payoff is structural. Three physical requirements — evolving for zero time does nothing, evolving in two stages equals evolving in one, and total probability stays 1 — force the operator to be unitary and its generator to be Hermitian. The Schrödinger equation becomes the infinitesimal fingerprint of those requirements, with the Hamiltonian revealed as the generator of time translation, exactly as momentum generates displacement. That is the viewpoint this chapter runs on: spectral decompositions, the Dyson series, the Schrödinger and Heisenberg pictures are all one operator $\hat U$ worn different ways. The completeness machinery is already paid for. Completeness Relation and Resolution of Identity assembled functions of operators from projectors and stopped, deliberately, at the eigenphases of $e^{-i\hat H t/\hbar}$ . This article cashes that promise: the phases become precession, oscillation, and the two-level beat it left waiting.

Key takeaways

$\hat U(t, t_0)$ propagates states: $|\psi(t)\rangle = \hat U(t, t_0)|\psi(t_0)\rangle$ — one operator, every initial condition.
Physics forces its shape: identity at $t = t_0$ , composition $\hat U(t_2, t_0) = \hat U(t_2, t_1)\hat U(t_1, t_0)$ , unitarity $\hat U^\dagger \hat U = \hat 1$ (probability conservation).
The infinitesimal step $\hat U(t_0 + dt, t_0) = \hat 1 - \tfrac{i}{\hbar}\hat H\,dt$ identifies $\hat H$ as the generator; it obeys the operator Schrödinger equation $i\hbar\,\partial_t \hat U = \hat H\hat U$ .
Unitarity $\Leftrightarrow$ Hermiticity of $\hat H$ — conservation of probability and reality of energy are the same fact.
Time-independent $\hat H$ : $\hat U = e^{-i\hat H(t - t_0)/\hbar}$ , acting on each energy eigenstate as a pure phase $e^{-iE_n t/\hbar}$ .
Three regimes: constant $\hat H$ (one exponential) · $[\hat H(t'), \hat H(t'')] = 0$ (integrate the exponent) · non-commuting (time-ordered exponential — Time-Ordered Exponential and the Dyson Series).
Key idea: all dynamics of a closed system is one unitary operator generated by the Hamiltonian — its unitarity is probability conservation, its composition law is the consistency of time itself.

The core insight

All quantum dynamics of a closed system is the action of one unitary operator $U(t,t_0)$ generated by the Hamiltonian via $i\hbar\,\partial_t U = H U$ ; its unitarity is conservation of probability and its group law $U(t_2,t_1)U(t_1,t_0)=U(t_2,t_0)$ is the consistency of time evolution.

Where this lives today

Every gate in a quantum computer is a hand-built $\hat U$ : a calibrated pulse making the qubit Hamiltonian act for exactly the right duration. Digital quantum simulation is the product-limit construction run in reverse — approximate $e^{-i\hat Ht/\hbar}$ for an intractable $\hat H$ by Trotterized slices of easy pieces, the $1/N$ error scaling of equation (2) setting the gate budget. And the 28-GHz-per-tesla precession of Example 1 is working hardware: electron spin resonance reads molecular environments off exactly that clock, while its nuclear cousin runs every MRI scanner. When this chapter later derives pictures, propagators, and perturbation series, each will be this one operator re-dressed.

The physical picture

Think of $\hat U(t, t_0)$ as a rigid rotation of the whole Hilbert space. A rotation preserves lengths and angles; $\hat U$ preserves norms and inner products — which is to say, probabilities. The Hamiltonian is the axis and rate of the rotation: just as momentum generates translations (the Weyl-form card of Canonical Commutation Relation), energy generates motion through time. States with definite energy are the directions the rotation does not mix — each spins in place at its own phase rate $E_n/\hbar$ — and all visible dynamics is interference between phases spinning at different rates.

Figure

One operator moves every state — U as a rigid rotation of Hilbert space, H as its generator

The second picture is composition. Evolving from 9:00 to 9:02 and then 9:02 to 9:05 must equal evolving from 9:00 to 9:05 — time evolution has no seams: $\hat U(t_2, t_0) = \hat U(t_2, t_1)\hat U(t_1, t_0)$ , the earlier interval acting first (rightmost). This unassuming rule, plus unitarity, is nearly the whole theory: it forces an infinitesimal-generator structure, and for constant $\hat H$ it upgrades the family to a one-parameter group — operators depending only on elapsed time, multiplying by adding arguments. Third: what makes the problem easy or hard is not whether $\hat H$ depends on time, but whether $\hat H$ at one time commutes with $\hat H$ at another. Constant Hamiltonians give a clean exponential. Time-dependent but self-commuting ones just integrate the exponent. Non-commuting ones are like composing rotations about different axes — order matters at every instant, and the honest bookkeeping is the time-ordered exponential built two articles ahead.

Figure

Which U for which H — the three-regime decision tree of quantum dynamics

The defining requirements — and the equation they force

Quantum dynamics asks for a family of operators doing three things. The contract:

\hat U(t_0, t_0) = \hat 1, \qquad \hat U(t_2, t_0) = \hat U(t_2, t_1)\,\hat U(t_1, t_0), \qquad \hat U^\dagger \hat U = \hat 1 \qquad (1)

where: the first clause says zero elapsed time changes nothing; the second is seamlessness (read right to left: earlier segment first); the third is unitarity — the inner-product-preserving class built at Unitary Operators and Transformations — demanded so a normalized state stays normalized forever. Linearity is also assumed: evolution must respect superpositions, or interference physics falls apart. From this contract alone, the dynamical law follows. That construction — Sakurai's opening move of quantum dynamics — is the first derivation.

Derivation AFrom three physical requirements to the operator Schrödinger equation

We derive the equation of motion for the evolution operator without assuming the Schrödinger equation — the requirements of contract (1) plus smoothness in time do all the work. The strategy: look at an infinitesimally short evolution, expand to first order, and let unitarity and composition dictate the structure of the linear term. What emerges is a Hermitian generator whose physical identity — the Hamiltonian — is fixed by the classical correspondence of energy with time translation, and an operator differential equation that contains the state-level Schrödinger equation as a corollary. Watch how little input produces how much output.

Step 1. Expand a short step. For an infinitesimal interval $dt$ , smoothness lets us write

\hat U(t_0 + dt, t_0) = \hat 1 - \frac{i}{\hbar}\,\hat\Omega\,dt + O(dt^2)

with some operator $\hat\Omega$ to be constrained. The factor $-i/\hbar$ is a convention that makes $\hat\Omega$ come out Hermitian with units of energy. Why start linear: the identity at $dt = 0$ is clause one of the contract; any smooth family departs from it linearly, and everything hinges on what the linear coefficient must be.

Step 2. Impose unitarity at first order:

\hat U^\dagger \hat U = \Big(\hat 1 + \frac{i}{\hbar}\hat\Omega^\dagger dt\Big)\Big(\hat 1 - \frac{i}{\hbar}\hat\Omega\,dt\Big) = \hat 1 + \frac{i}{\hbar}\big(\hat\Omega^\dagger - \hat\Omega\big)\,dt + O(dt^2)

For this to equal $\hat 1$ at order $dt$ , we need $\hat\Omega^\dagger = \hat\Omega$ : the generator must be Hermitian. An observable has appeared out of pure bookkeeping. Why this is remarkable: probability conservation alone manufactures a Hermitian operator — the mirror-symmetric class of Hermitian Operators and Observables — before any physics names it.

Step 3. Name the generator. Which observable generates time translation? Classical mechanics answers: the Hamiltonian generates time evolution, as momentum generates displacement — and the translation-operator analogy $e^{-ia\hat p/\hbar}$ (the Weyl card of Canonical Commutation Relation) makes $\hbar$ 's bookkeeping match. Set $\hat\Omega = \hat H$ , the energy operator. This identification is a physical postulate — the same one Time-Dependent Schrödinger Equation makes at the state level — judged by its scorecard, not derived.

Step 4. Build the differential equation from composition. Compose a finite evolution with one more infinitesimal step:

\hat U(t + dt, t_0) = \hat U(t + dt, t)\,\hat U(t, t_0) = \Big(\hat 1 - \frac{i}{\hbar}\hat H(t)\,dt\Big)\hat U(t, t_0)

Subtract $\hat U(t, t_0)$ , divide by $dt$ , and take the limit:

i\hbar\,\frac{\partial}{\partial t}\hat U(t, t_0) = \hat H(t)\,\hat U(t, t_0), \qquad \hat U(t_0, t_0) = \hat 1

Why this contains the state equation: apply both sides to a fixed $|\psi(t_0)\rangle$ and the operator equation becomes $i\hbar\,\partial_t|\psi(t)\rangle = \hat H|\psi(t)\rangle$ — the ket-space law of the previous article, now recognized as the infinitesimal fingerprint of unitary, seamless evolution. *Verified: the exponential solution satisfies the operator equation symbolically (sympy, 2×2 $\hat H$ : $i\,d\hat U/dt - \hat H\hat U = 0$ exactly), and the composition law $\hat U(1.9) = \hat U(0.7)\hat U(1.2)$ holds to $10^{-12}$ on a random 3×3 Hermitian Hamiltonian (verify script, 20 checks green).*

Unitarity and Hermiticity are the same fact

The construction above went from unitarity to a Hermitian generator. The traffic runs both ways, and the round trip is worth one compact derivation because it is the exam-grade statement of "probability is conserved."

Go deeper · Derivation B · 1 minHermiticity of H is exactly unitarity of U

Derivation BHermiticity of H is exactly unitarity of U

We prove that $\hat U^\dagger\hat U$ is constant in time if and only if $\hat H = \hat H^\dagger$ , so the initial condition $\hat U(t_0, t_0) = \hat 1$ pins $\hat U^\dagger\hat U = \hat 1$ forever. The strategy is to differentiate the product $\hat U^\dagger\hat U$ and watch the two equations of motion — one for $\hat U$ , one for its adjoint — cancel precisely when the Hamiltonian equals its own adjoint. This is the operator-level twin of the wave-mechanics norm-conservation proof, which lives at Conservation of Normalization and is recalled, not re-run, here.

Step 1. Take the adjoint of the operator Schrödinger equation. Using the product-reversal rule of Hermitian Conjugate and Adjoint Operators:

i\hbar\,\partial_t \hat U = \hat H\,\hat U \quad\Longrightarrow\quad -i\hbar\,\partial_t \hat U^\dagger = \hat U^\dagger\,\hat H^\dagger

Step 2. Differentiate the product and insert both equations:

\partial_t\big(\hat U^\dagger\hat U\big) = (\partial_t\hat U^\dagger)\,\hat U + \hat U^\dagger\,(\partial_t\hat U) = \frac{i}{\hbar}\,\hat U^\dagger\big(\hat H^\dagger - \hat H\big)\hat U

Why the structure is decisive: the derivative of the product is sandwiched between $\hat U^\dagger$ and $\hat U$ , so it vanishes for every state exactly when $\hat H^\dagger - \hat H = 0$ — no cancellation is available from anything else.

Step 3. Read both directions. If $\hat H = \hat H^\dagger$ : the derivative vanishes, $\hat U^\dagger\hat U$ keeps its initial value $\hat 1$ , and every norm $\langle\psi(t)|\psi(t)\rangle = \langle\psi(t_0)|\hat U^\dagger\hat U|\psi(t_0)\rangle$ is frozen — probability conserved. If instead $\hat H$ has an anti-Hermitian part $-i\Gamma/2$ : the same algebra gives exponential norm decay — precisely the optical-potential model of Conservation of Normalization, whose $\tau = \hbar/\Gamma$ lifetime is the controlled way to break this theorem on purpose. Verified: numerically — $\|\hat U^\dagger\hat U - \hat 1\| < 10^{-12}$ and norm conservation to $10^{-14}$ for a random 3×3 Hermitian $\hat H$ ; inverse = adjoint ( $\hat U(-t) = \hat U(t)^\dagger$ ) to $10^{-12}$ (verify script).

Solving the contract: the exponential and its spectral form

For a time-independent Hamiltonian the operator equation is solved by compounding the infinitesimal step $N$ times over the interval and letting $N \to \infty$ — legal because every factor involves the same $\hat H$ and commuting factors multiply like numbers:

\hat U(t, t_0) = \lim_{N\to\infty}\Big(\hat 1 - \frac{i}{\hbar}\hat H\,\frac{t - t_0}{N}\Big)^{\!N} = \exp\!\Big[-\frac{i}{\hbar}\hat H\,(t - t_0)\Big] \qquad (2)

where: the exponential means its power series, and the limit is the operator version of $(1 + x/N)^N \to e^x$ . The verify script's error ladder makes the convergence concrete: truncating at $N = 10, 100, 1000$ steps leaves errors $0.24, 0.022, 0.0022$ — shrinking like $1/N$ , which is also why product-formula ("Trotterized") quantum simulation works. Differentiating (2) returns the operator Schrödinger equation, closing the loop. How does the exponential act? Completeness Relation and Resolution of Identity assembled functions of an operator from its spectral projectors — eigenvalues go through the function, eigenvectors stand still — and applied it to $e^{-i\hat Ht/\hbar}$ , stopping at the eigenphases. Cash the promise:

\hat U(t, 0) = \sum_n e^{-iE_n t/\hbar}\,|E_n\rangle\langle E_n| \qquad (3)

where: each energy eigenstate is multiplied by a pure phase turning at rate $E_n/\hbar$ — the spectral form of time evolution. A single eigenstate goes nowhere observable (the stationary-state story, next articles); superpositions evolve only through phase differences. Equation (3) is verified against the direct matrix exponential to $10^{-12}$ on the two-level system of Eigenvalues and Eigenstates — and Example 2 below turns its eigenphase drift into a real oscillation. The full craft of using (3) — the expand–rotate–resum algorithm, beats, revivals — is the next-but-two article, Spectral Decomposition of Time Evolution. One honest caveat with a famous name. In infinite dimensions, "Hermitian generator" needs the self-adjoint fine print flagged at Hermitian Operators and Observables — and the rigorous statement is Stone's theorem: every strongly continuous one-parameter unitary group is $e^{-i\hat At/\hbar}$ for a unique self-adjoint $\hat A$ , and conversely. Self-adjointness is exactly the credential that buys a Hamiltonian its evolution operator. Proving that real atoms hold this credential was a landmark — the portrait below.

Which U for which Hamiltonian

The taxonomy every dynamics problem starts with:

regime	evolution operator	why
$\hat H$ constant	$\exp[-\tfrac{i}{\hbar}\hat H(t - t_0)]$	one generator; the product limit (2) applies verbatim
$\hat H(t)$ , but $[\hat H(t'), \hat H(t'')] = 0$	$\exp[-\tfrac{i}{\hbar}\int_{t_0}^t \hat H(t')\,dt']$	commuting generators — exponents add like numbers
$[\hat H(t'), \hat H(t'')] \neq 0$	$\mathcal T\exp[-\tfrac{i}{\hbar}\int_{t_0}^t \hat H(t')\,dt']$	order matters at every instant; see Time-Ordered Exponential and the Dyson Series
any regime, infinitesimal	$\hat 1 - \tfrac{i}{\hbar}\hat H(t)\,dt$	the generator statement itself — always true

Row three is not the naive integrated exponential — that formula is wrong whenever $\hat H$ fails to commute with itself across times, and the failure is quantitative (a two-piece counterexample is worked to a number in the Dyson article). The composition law survives in every regime; only time-translation invariance — dependence on $t - t_0$ alone — is special to constant $\hat H$ .

Limiting cases

case	result	meaning
$t \to t_0$	$\hat U \to \hat 1 - \tfrac{i}{\hbar}\hat H\,(t - t_0)$	the generator is the short-time fingerprint
eigenstate input	$\hat U\|E_n\rangle = e^{-iE_n t/\hbar}\|E_n\rangle$	pure phase — stationary states, next article
$[\hat A, \hat H] = 0$	$\hat U^\dagger\hat A\hat U = \hat A$ , $\langle\hat A\rangle$ frozen	constants of the motion — the conservation story ahead
$\hat H \to \hat H + c$	$\hat U \to e^{-ict/\hbar}\,\hat U$	global phase only — the energy-shift theorem of the previous article
run time backward	$\hat U(t_0, t) = \hat U^\dagger(t, t_0)$	unitary evolution is reversible; nothing is lost between measurements
$\hbar \to 0$ at fixed energies	phases spin infinitely fast	classical limit lives in stationary-phase cancellations (propagator articles ahead)

Common trapFor a time-dependent Hamiltonian, U equals the exponential of the integral of H

Only if the Hamiltonian commutes with itself at all pairs of times, $[\hat H(t'), \hat H(t'')] = 0$ — for example, a field that changes strength but never direction. The moment the direction of $\hat H$ in operator space turns (a rotating drive field, a gate sequence alternating $\sigma_z$ - and $\sigma_x$ -type couplings), the integrated exponential $\exp[-\tfrac{i}{\hbar}\int \hat H\,dt']$ is simply the wrong operator: composing evolutions about different axes depends on their order, exactly as rotations do. The honest object is the time-ordered exponential, and the error of ignoring the ordering is a computable, first-order-visible gap — Time-Ordered Exponential and the Dyson Series constructs it and puts a number on the failure. On an exam: quote the exponential only after checking the self-commutation box.

Worked examples

Example 1 — spin precession, the cleanest U in physics (confidence builder). An electron sits in a uniform field $\mathbf B = B\hat z$ with $B = 1$ T. Its Hamiltonian is $\hat H = \omega \hat S_z$ with $\omega = g_e \mu_B B/\hbar$ . Find $\omega$ , write $\hat U(t)$ , and evolve the state $|{+x}\rangle$ .

Step 1. The rate. $\omega = g_e\mu_B B/\hbar = \dfrac{2.0023 \times 9.274\times10^{-24} \times 1}{1.055\times10^{-34}} = \boxed{1.761\times10^{11}\ \mathrm{rad/s}}$ — i.e. $f = \omega/2\pi = \boxed{28.02\ \mathrm{GHz}}$ , the electron-spin-resonance frequency at one tesla.

Step 2. The operator. Since $\hat S_z = (\hbar/2)\sigma_z$ : $\hat U(t) = e^{-i\omega t\,\sigma_z/2}$ — a rotation about $\hat z$ by angle $\omega t$ , diagonal in the $|\pm z\rangle$ basis with eigenphases $e^{\mp i\omega t/2}$ . The spectral form (3) with two terms.

Step 3. Evolve $|{+x}\rangle = (|{+z}\rangle + |{-z}\rangle)/\sqrt2$ . Each branch takes its own phase; the relative phase $e^{-i\omega t}$ is the physics: $\langle\sigma_x\rangle(t) = \cos\omega t$ , $\langle\sigma_y\rangle(t) = +\sin\omega t$ (machine-checked at 41 times) — the spin expectation precesses in the $xy$ -plane at $\omega$ , one full turn every $T = 2\pi/\omega = \boxed{35.7\ \mathrm{ps}}$ .

Step 4. A quantum signature for free: at $t = T$ the expectations return, but $\hat U(T) = e^{-i\pi\sigma_z} = -\hat 1$ — the spinor comes back with a minus sign (verified exactly). A $2\pi$ rotation is not the identity on spin- $\frac12$ ; the sign is physical in interferometry, and its full story belongs to the angular-momentum chapter's SU(2) article.

Example 2 — cashing the eigenphase handshake: the two-level oscillation (exam level). Completeness Relation and Resolution of Identity built $e^{-i\hat Ht/\hbar}$ for the coupled two-level system of Eigenvalues and Eigenstates — $\varepsilon_1 = 1.0$ , $\varepsilon_2 = 2.0$ , coupling $g = 0.3$ eV, eigenvalues $E_\mp = 0.917 / 2.083$ eV — and reported the eigenphase beat time $2\pi\hbar/(E_+ - E_-) = 3.55$ fs, promising this article would turn it into oscillation. Prepare the system in the bare state $|1\rangle$ and find the probability of finding it in $|2\rangle$ at time $t$ .

Step 1. Expand the initial state over eigenstates (mixing angle $\tan 2\theta = 2g/(\varepsilon_2 - \varepsilon_1) \Rightarrow \theta = 15.5°$ ), apply (3), resum. The off-diagonal element of $\hat U$ gives, exactly,

P_{1\to2}(t) = \sin^2 2\theta\,\sin^2\!\Big(\frac{(E_+ - E_-)\,t}{2\hbar}\Big)

verified against the direct matrix exponential to $10^{-12}$ across two full periods.

Step 2. Read the two factors. The amplitude $\sin^2 2\theta = \boxed{0.265}$ is set by geometry — how far the eigenbasis is rotated from the bare basis. The clock runs at the Bohr frequency: full oscillation period $= 2\pi\hbar/(E_+ - E_-) = \boxed{3.55\ \mathrm{fs}}$ , the promised beat.

Step 3. The peak transfer: $P_{\max} = 26.5\%$ at $t = \boxed{1.77\ \mathrm{fs}}$ (half the beat period), then the population flows back — forever, since evolution is unitary and reversible. Complete transfer would need $\theta = 45°$ : degenerate bare levels ( $\varepsilon_1 = \varepsilon_2$ ), the resonance condition. This formula is the Rabi oscillation of a static coupling; the driven version rides on the same algebra in the time-dependent chapter.

Step 4. Conservation check, both flavors: $P_{1\to1} + P_{1\to2} = 1$ at every instant (unitarity), and $\langle\hat H\rangle$ is constant while $\langle\sigma_x\rangle$ oscillates — $[\hat H, \hat H] = 0$ but $[\sigma_x, \hat H] \neq 0$ (both machine-checked), previewing the constants-of-the-motion criterion.

The man who proved the hydrogen atom evolves

Origins · 1 minTosio Kato and how this was discovered

Tosio Kato

1917–1999 · University of Tokyo, then Berkeley

Stone's theorem says a self-adjoint Hamiltonian generates a unitary group — but for twenty years after von Neumann sharpened "Hermitian" into "self-adjoint," nobody knew whether the Hamiltonian of a real atom, Coulomb singularities and all, actually held that credential. If not, $e^{-i\hat Ht/\hbar}$ for hydrogen would be formalism without a referent. Kato settled it in 1951: the Coulomb Hamiltonians of atoms and molecules are essentially self-adjoint — the singular attraction is dominated by the kinetic term (the Kato–Rellich smallness argument), so the evolution operator of every atom in the universe rigorously exists. He had proved it years earlier in wartime Tokyo, working while evacuated to the countryside with tuberculosis, unable to publish; the paper appeared in an American journal after a colleague circulated it. Kato's Perturbation Theory for Linear Operators (1966) became the standard reference for how spectra and evolution respond when Hamiltonians change; the adiabatic theorem in modern operator form is his too (1950). A quiet figure who loved botany and precise estimates, he supplied quantum dynamics with the thing physicists rarely notice needing: the proof that its central object exists. (Von Neumann's self-adjointness fine print lives at Hermitian Operators and Observables — this band is the existence-of-dynamics story only.)

History

1926

— Schrödinger's equation gives dynamics as a differential law; solving it state by state is the only tool

1927

— Weyl's group-theoretic quantum mechanics recasts evolution as a one-parameter unitary family

1932

— Stone proves the theorem pairing unitary groups with self-adjoint generators; von Neumann's Grundlagen frames dynamics as U(t)

1949

— Dyson's series organizes the time-ordered exponential for QED's interaction picture

1951

— Kato proves atomic Hamiltonians self-adjoint: the hydrogen atom's U(t) rigorously exists

1996

— Lloyd's universal quantum simulation runs on Trotterized products of short-time evolution operators

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., Sec. 2.1–2.2 · Griffiths, Introduction to Quantum Mechanics, 3rd ed., Ch. 2 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 4 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Ch. III · Townsend, A Modern Approach to Quantum Mechanics, 2nd ed., Ch. 4.

Provenance. Definitions and properties (operator Schrodinger equation, unitarity, composition law, exponential and time-ordered forms) follow Sakurai & Napolitano Modern QM 2nd ed. Sec. 2.1-2.2, Shankar Ch. 4, Cohen-Tannoudji Vol. I Ch. III, Griffiths 3rd ed. Ch. 2. Verifier should: (1) confirm $\partial_t(U^\dagger U)=0$ algebra uses only $H=H^\dagger$ ; (2) check the $(1+x/N)^N\to e^x$ operator limit requires commuting factors; (3) verify the spectral-form resummation; (4) confirm the infinite-square-well revival time $T_{rev}=4mL^2/(\pi\hbar)$ from $E_n=n^2\pi^2\hbar^2/(2mL^2)$ and the Rabi population $\sin^2(\Omega_R t/2)$ . All numbers are standard textbook values. For this edition all quantitative claims were machine-verified (numpy/scipy/sympy, 20 checks green): the operator-ODE and composition/inverse/unitarity identities; the product-limit error ladder ( $0.24/0.022/0.0022$ at $N = 10/10^2/10^3$ ); the spectral assembly against the matrix exponential ( $10^{-12}$ ); the precession numbers ( $\omega = 1.761\times10^{11}$ rad/s, $f = 28.02$ GHz, $T = 35.7$ ps, $\langle\sigma_{x,y}\rangle$ traces, the $2\pi$ spinor sign); and the two-level oscillation ( $E_\mp = 0.917/2.083$ eV, $\theta = 15.5°$ , amplitude $0.265$ , beat $3.55$ fs, peak at $1.77$ fs). Per the chapter ownership map: the state-level equation anatomy belongs to Time-Dependent Schrodinger Equation (previous serial); the expand–rotate–resum craft, beats, and the square-well revival ( $T_{rev} = 4mL^2/\pi\hbar$ , quoted from the base) to Spectral Decomposition of Time Evolution; the time-ordered exponential and its counterexample number to Time-Ordered Exponential and the Dyson Series; the rank-1 spectral assembly consumed in equation (3) to Completeness Relation and Resolution of Identity (its 3.55 fs eigenphase handshake is cashed in Example 2); the unitary class and generator doctrine to Unitary Operators and Transformations; conservation-law development to Constants of the Motion and Conservation Laws; the propagator kernel $\langle x|\hat U|x'\rangle$ to The Quantum Propagator.

Builds on: Time-Dependent Schrodinger Equation · Stationary States and Energy Eigenstates · The Schrodinger Picture

Leads to: Spectral Decomposition of Time Evolution · Time-Ordered Exponential and the Dyson Series · The Heisenberg Picture · Equivalence of the Dynamical Pictures · Constants of the Motion and Conservation Laws

Is U(t,t₀) something I measure, or just bookkeeping?

Its matrix elements are measured constantly: $|\langle f|\hat U(t)|i\rangle|^2$ is the transition probability from $|i\rangle$ to $|f\rangle$ — Example 2's 26.5% is exactly $|U_{21}|^2$ at its peak, and quantum-computing labs reconstruct entire gate unitaries element by element (process tomography). What you never measure is $\hat U$ 's overall phase — consistent with the energy-shift theorem: adding a constant to $\hat H$ re-phases $\hat U$ without touching any probability.

Why does U sit to the LEFT in the composition law, with the earliest interval rightmost?

Operators act on kets from the left: in $\hat U(t_2,t_1)\hat U(t_1,t_0)|\psi(t_0)\rangle$ the state meets the $t_0 \to t_1$ segment first, as chronology demands. Misordering is harmless for constant $\hat H$ but wrong the moment the Hamiltonian's direction turns — the same order-sensitivity that forces time-ordering in the Dyson series. Reading products right-to-left is the most protective habit in this chapter.

Does U exist when the Hamiltonian depends on time? The exponential formula seems to die.

The formula (2) dies; the operator survives. The defining contract never referenced constancy of $\hat H$ . For self-commuting time dependence the exponent just integrates; in general the solution is the time-ordered exponential, built two articles ahead. Losing the simple formula costs only time-translation invariance ( $\hat U$ now depends on both endpoints, not the difference).

Is the Hamiltonian the only operator that could generate time evolution?

Unitarity forces some Hermitian generator (Derivation A, Step 2); which observable it is, is physics. The identification with energy comes from correspondence: classically the Hamiltonian generates time translation through Poisson brackets, and the momentum-generates-displacement analogy fixes the pairing and the $\hbar$ . The deep version — time-translation symmetry pairs with energy conservation — is Noether's logic, developed at this chapter's symmetry articles.

If unitary evolution is reversible, where does irreversibility in the real world come from?

Not from $\hat U$ — a closed system's evolution can always be undone by $\hat U^\dagger$ (spin echoes do exactly this, rewinding dephased magnetization). Irreversibility enters two ways: measurement (the projection postulate is not unitary — the collapse story of Measurement and Wavefunction Collapse), and openness — a system entangling with an unmonitored environment has no unitary of its own. Inside this chapter, everything is closed and reversible by construction.

My textbook writes U(t) = exp(−iHt/ħ) with t₀ = 0 everywhere. When does the t₀ matter?

For constant $\hat H$ , never — only elapsed time enters, and setting $t_0 = 0$ is free. For time-dependent Hamiltonians the family genuinely depends on both arguments: a pulse applied at 9:00 and the same pulse applied at 9:05 produce different evolutions relative to the lab clock, and $\hat U(t_2, t_1) \neq \hat U(t_2 - t_1, 0)$ in general. Keeping both slots is the notational tell that you are ready for driven systems.