The Heisenberg Picture

Quantum field theory never left this picture: a field $\hat\phi(\mathbf x,t)$ is a Heisenberg operator labeled by spacetime, equal-time commutation relations are equation (5) writ large, and the rigidity proved here is why they can be postulated once and trusted forever. Two-time correlation functions — $\langle\hat x_H(t)\hat x_H(0)\rangle$ in condensed matter, $\langle\hat E(t)\hat E(0)\rangle$ in photon-correlation spectroscopy and input–output theory — are native Heisenberg objects with no single-clock Schrödinger translation. The sharpest modern twist is computational. Gottesman's 1998 "Heisenberg representation of quantum computers" showed that for Clifford circuits one can classically track how a small set of operators (stabilizers) conjugates through the circuit instead of the exponentially large state — a $2^n$-dimensional nightmare turned polynomial. Stabilizer simulators and most of quantum error-correction theory run in this picture; the hardware itself runs prepare–evolve–measure in the previous article's. Modern quantum computing is, operationally, a dialogue between the two cameras.

Ride the merry-go-round instead of watching it. In the Schrödinger picture you stand in the laboratory: the state-arrow spins, the operator axes stay bolted down. The Heisenberg picture straps you to the arrow. From your seat the state appears motionless — it is your reference — and the laboratory of measuring axes wheels around you in the opposite sense. Same relative geometry instant by instant, same measured numbers; opposite assignment of who moves.

The cartoon earns its keep on the subtlety that trips even good students: in the Heisenberg picture the eigenkets of observables move backward. If the operator dresses as $\hat U^\dagger\hat A\hat U$, its eigenkets at time $t$ are $\hat U^\dagger|a\rangle$ — the conjugate mirror of the Schrödinger state's $\hat U|\psi\rangle$. The eigenvalues ride along unchanged (a similarity transformation cannot move a spectrum), but the axes sweep opposite to the state they used to watch. The who-carries-time ledger below is this article's contribution to the pictures trilogy; the previous article's decision table says when to prefer which camera, and we do not reprint it.

Demand only that physics not change. The previous article regrouped the amplitude sandwich once; we now promote the regrouping to a definition:

where: $\hat U$ is the same time-evolution operator built in The Time-Evolution Operator — the picture invents no new dynamics, it re-drapes the old one. At $t_0$ the pictures coincide: one set of operators, one state, two future bookkeepings. The frozen state is defined by undoing the motion, $|\psi\rangle_H = \hat U^\dagger|\psi(t)\rangle_S = |\psi(t_0)\rangle_S$, so that the sandwich closes:

where: the equality is associativity, verified to $10^{-13}$ on random systems — the pictures were never different calculations, only different groupings. For time-independent $\hat H$ the dressing is $\hat U = e^{-i\hat Ht/\hbar}$, and one consequence deserves its own line: $\hat H$ commutes with its own exponential, so $\hat H_H = \hat U^\dagger\hat H\hat U = \hat H$ — the same operator in both pictures. (With explicit time dependence this fails; the pitfall block returns to it.) Differentiating equation (1) with $i\hbar\,\partial_t\hat U = \hat H\hat U$ produces the celebrated operator law of motion,

where: the commutator term is the dynamical flow; the second passes through any hand-built explicit time dependence. Equation (3) — its derivation, its Poisson-bracket correspondence $\{A,H\} \leftrightarrow [\hat A,\hat H]/i\hbar$, its applications from the oscillator's operator trajectories to conservation laws — is the entire business of the next article, The Heisenberg Equation of Motion; this article deliberately stops at the statement. What we can harvest without the equation is remarkable: the free particle falls to bare algebra below, because conjugation by an exponential sometimes has a terminating series.

The picture's kinematic backbone: unitary conjugation preserves every algebraic relation among operators — products conjugate to products, sums to sums, hence commutators to commutators:

where: the identity is two insertions of $\hat U\hat U^\dagger = \hat 1$ (checked numerically to $10^{-13}$). Apply it to the canonical pair with the relation $[\hat x,\hat p] = i\hbar$ owned by Canonical Commutation Relation: the right side conjugates a multiple of the identity, which no unitary can touch, so

where: the equal-time commutation relations are time-translation invariant. The kinematics is rigid; only the time labels move — which is why quantum field theory can postulate equal-time relations for Heisenberg fields and never worry that dynamics erodes them. Note what is not claimed: operators at different times need not commute — the free particle's $[\hat x_H(t), \hat x_H(0)]$ below is famously nonzero, and that failure is physics, not pathology.

Sandwich a Heisenberg operator between energy eigenkets $\hat H|n\rangle = E_n|n\rangle$ and the dressing exponentials turn into pure phases:

where: every matrix element is a clock running at a Bohr frequency — diagonal entries frozen, off-diagonal entries spinning at exactly the transition frequencies an atom radiates. This is the 1925 theory verbatim: Heisenberg's arrays were these oscillating entries, written down before anyone knew what they were matrices of. Spectroscopy reads (6) directly — spectral lines are the clock rates of the dipole operator's off-diagonal elements.

Equation (6) also makes conservation laws visible by inspection: if $[\hat H,\hat A] = 0$ (and no explicit clock), every surviving matrix-element clock has $\omega_{nm} = 0$ and $\hat A_H(t) = \hat A_H(0)$ — a constant of the motion, the theme Constants of the Motion and Conservation Laws develops. The verifier exhibits both facts: sixteen matrix elements carrying their Bohr phases exactly, and a polynomial in $\hat H$ refusing to move.

Example 1 — how far does the operator drift? (confidence builder). An electron beam has mean velocity $\langle\hat p\rangle/m = 1.0\times10^{6}$ m/s. Using the Heisenberg solution, how far does the position operator's expectation drift in 1.0 ns, for any state with that mean momentum?

Step 1. The operator solution is state-independent: $\hat x_H(t) = \hat x + (\hat p/m)t$, so $\langle\hat x\rangle_t = \langle\hat x\rangle_0 + (\langle\hat p\rangle/m)\,t$.

Step 2. Numbers: $\Delta\langle\hat x\rangle = 10^{6} \times 10^{-9} = \boxed{1.0\ \text{mm}}$ — for every state sharing that mean momentum, Gaussian or grotesque. The picture solved the dynamics once, for all states simultaneously; that is what "the operator carries the dynamics" buys. (The wave-register cousin is Ehrenfest's Theorem's $d\langle x\rangle/dt = \langle p\rangle/m$; the general commutator engine is this chapter's own Ehrenfest's Theorem, four serials ahead.)

Example 2 — the two-time commutator forces spreading (exam level). An electron is prepared with position spread $\sigma_{x}(0) = 1.0$ nm. Use $[\hat x_H(t),\hat x_H(0)] = -i\hbar t/m$ to bound $\sigma_x$ at $t = 1.0$ ps, and compare with the exact Gaussian law.

Step 1. The Robertson bound (cited to Generalized Uncertainty Principle): $\sigma_{x(t)}\,\sigma_{x(0)} \geq \tfrac12|\langle[\hat x_H(t),\hat x_H(0)]\rangle| = \hbar t/2m$.

Step 2. Numbers: $\sigma_x(t) \geq \dfrac{\hbar t}{2m\sigma_x(0)} = \dfrac{1.0546\times10^{-34}\times10^{-12}}{2\times9.109\times10^{-31}\times10^{-9}} = \boxed{57.9\ \text{nm}}$. A nanometer-sharp electron cannot stay sharp — blurring by a factor ≥ 58 in one picosecond is an algebraic obligation, not a dynamical accident.

Step 3. Compare the exact law owned by Wave Packet Spreading and Dispersion: $\sigma(t) = \sigma_0\sqrt{1 + (\hbar t/2m\sigma_0^2)^2}$. With $\hbar t/2m\sigma_0^2 = 57.9$: $\sigma(t) = \boxed{57.9\ \text{nm}}$ — the Gaussian packet saturates the two-time bound asymptotically (ratio 1.0001, machine-checked). The minimum-uncertainty packet spreads exactly as slowly as the algebra permits; the Heisenberg picture found the floor in one commutator.

Example 3 — the qubit revisited, camera two (exam level). Rerun the previous article's 5 GHz qubit ($\hat H = \tfrac{\hbar\omega}{2}\hat\sigma_z$, state $|{+}x\rangle$) in the Heisenberg picture: find $(\hat\sigma_x)_H(t)$ and the readings at $t = 50$ ps.

Step 1. Conjugate: $(\hat\sigma_x)_H(t) = e^{i\omega t\hat\sigma_z/2}\,\hat\sigma_x\,e^{-i\omega t\hat\sigma_z/2} = \cos\omega t\,\hat\sigma_x - \sin\omega t\,\hat\sigma_y$ (machine-checked at three times). The operator frame rotates with the $-\sin$ signature — backward relative to the Schrödinger state's $+x \to +y$ precession, exactly as the ledger demands.

Step 2. Sandwich in the frozen $|{+}x\rangle$: $\langle(\hat\sigma_x)_H\rangle_t = \cos\omega t$, $\langle(\hat\sigma_y)_H\rangle_t = +\sin\omega t$. At $t = 50$ ps (quarter period): $\boxed{0}$ and $\boxed{1}$ — identical to the Schrödinger camera, as equation (2) guarantees.

Step 3. The moral: the backward rotation lives in the operator components (the $-\sin\omega t\,\hat\sigma_y$ term), never in the measured values. If your two cameras ever disagree on a number, you have double-counted a dagger.

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

Provenance. Equations are standard QM textbook results. A verifier should recompute: (1) $A_H=U^\dagger A_S U$ and the Heisenberg EOM by differentiating with $i\hbar\dot U=HU$ (Sakurai §2.2); (2) free-particle and SHO Heisenberg solutions using $[x,p]=i\hbar$ (Shankar Ch. 4); (3) Bohr-frequency matrix elements $\omega_{nm}=(E_n-E_m)/\hbar$; (4) preservation of the canonical commutator under unitary conjugation. Cross-check with Cohen-Tannoudji Vol. I Complement G-III and Townsend Ch. 4. For this edition, items (1), (3), (4) plus every numeric claim were machine-verified (sympy/numpy, 24 checks): the terminating BCH ladder and free-particle solutions (symbolic and in a 200-level truncation, interior block, better than $10^{-13}$), the two-time commutator, CCR rigidity, backward base-ket motion with spectrum invariance, the Bohr-phase matrix elements, the constant-of-motion freeze, the qubit's frame rotation with picture-matched readings, and the 57.9 nm bound saturated by the exact Gaussian law. Item (2)'s SHO solution is deliberately NOT reproduced: per the chapter ownership map it belongs to The Heisenberg Equation of Motion, which owns the EOM and its applications — this article stops at the free particle, solvable by pure conjugation. The CCR and operator-expansion lemma are owned by Canonical Commutation Relation; the spreading law by Wave Packet Spreading and Dispersion; the equivalence theorem by Equivalence of the Dynamical Pictures.

Builds on: The Time-Evolution Operator · The Schrodinger Picture · Time-Dependent Schrodinger Equation

Leads to: The Heisenberg Equation of Motion · The Interaction (Dirac) Picture · Ehrenfest's Theorem · Constants of the Motion and Conservation Laws · Equivalence of the Dynamical Pictures