Constants of the Motion and Conservation Laws

Stationary States and Energy Eigenstates ended with a distinction it deliberately left dangling: a stationary state — one frozen ket, every observable's statistics stilled — is not the same thing as a conserved observable — one quantity whose statistics are stilled in every state. This article is the promised theory of the second thing. The criterion is a single commutator. If $[\hat{A},\hat{H}]=0$ and $\hat{A}$ carries no explicit clock, then $\hat{A}$ is a constant of the motion — and the conservation is far stronger than the obvious statement about averages. The Ehrenfest engine of the previous article makes $\langle\hat{A}\rangle$ constant in one line; this article proves the distribution-level theorem: every moment, and every individual outcome probability $P(a_n)$ , is frozen for every initial state. Whatever the measurement statistics of a conserved quantity look like at $t=0$ , they look identical forever. That is what "good quantum number" means, and why labels like $n$ , $\ell$ , parity survive dynamics as bookkeeping you can trust. Around the criterion grows an algebra with classical roots. Poisson knew in 1809 that the bracket of two conserved quantities is conserved; Jacobi proved it properly, and his identity — inherited by the commutator — makes the quantum version a two-line theorem: constants of the motion close under commutation. Conserved quantities also refuse to mix energy eigenspaces (a no-mixing lemma with selection-rule consequences), and by the compatibility theorem they join $\hat{H}$ in a common eigenbasis — the CSCO story, now with dynamics attached. A machine-run double well watches parity hold still to $10^{-13}$ while the packet sloshes; the question this article never answers — where do constants come from? — is exactly the next one's.

Key takeaways

Constant of the motion: $[\hat{A},\hat{H}]=0$ and $\partial_t\hat{A}=0$ . Then $\langle\hat{A}\rangle$ is frozen (Ehrenfest one-liner) — and much more.
THE theorem: every outcome probability $P(a_n,t)$ and every moment $\langle\hat{A}^k\rangle$ is time-independent, for every initial state — conservation at the level of full statistics.
Mechanism: $[\hat{A},\hat{H}]=0\Rightarrow[\hat{P}_n,\hat{U}(t)]=0$ — the eigenprojectors of a conserved observable commute with the evolution operator itself.
Good quantum numbers: eigenvalue labels of conserved observables survive evolution; a state born with $a_n$ stays in that eigenspace forever.
Jacobi closure: if $\hat{A}$ and $\hat{B}$ are constants, so is $\frac{1}{i}[\hat{A},\hat{B}]$ — constants form an algebra (Poisson's theorem, quantized).
No-mixing lemma: $[\hat{A},\hat{H}]=0$ forces $\langle a'|\hat{H}|a\rangle=0$ between different eigenvalues — $\hat{H}$ is block-diagonal in conserved labels.
Stationary STATE ≠ conserved OBSERVABLE: one frozen ket versus one quantity frozen in all kets — the contrast table below cashes the stationary-states article's deferral.
Key idea: commuting with the Hamiltonian freezes an observable's entire probability distribution, in every state, forever — conservation is a statement about statistics, not just averages.

The core insight

An observable with $[\hat{A},\hat{H}]=0$ and no explicit time dependence is a constant of the motion: its eigenprojectors commute with $\hat{U}(t)$ , so every outcome probability and every moment of $\hat{A}$ is frozen in every state — and its eigenvalues are good quantum numbers that survive dynamics.

Where this lives today

Good quantum numbers are the working currency of spectroscopy and quantum computing alike. Atomic physicists trust the address $|n,\ell,m,s\rangle$ because each slot commutes with the atomic Hamiltonian to excellent approximation — and they diagnose the violations (fine structure, external fields) by watching which labels stop being conserved and which survive. Superconducting-qubit designers run the same logic in reverse: a "quantum-nondemolition" readout is engineered by making the measured observable commute with the system–probe Hamiltonian, so that this article's frozen-distribution theorem protects the qubit's census while information leaks to the amplifier. At the frontier, the failure of the harvest is a field of its own: generic interacting many-body systems have essentially no local constants of the motion beyond energy — the foundation of thermalization — while many-body-localized and integrable systems hoard extensive families of them and refuse to thermalize. "Count the conserved quantities" has become a sharp experimental question about cold-atom chains, answered by watching exactly the kind of frozen censuses this article's verifier stages in miniature.

The physical picture

Think of Hilbert space as a country and the conserved observable's eigenspaces as its provinces. The theorem of this article says the dynamics respects the borders: $\hat{U}(t)$ may swirl population within each province however it likes, but not a drop of probability ever crosses a border. A state born 30% in province $a_1$ and 70% in $a_2$ keeps that split to the end of time. That is conservation, quantum style — not a frozen number but a frozen census.

Figure

Conserved observables partition Hilbert space into provinces the dynamics cannot cross

Hold this against the stationary-state picture and the deferred distinction snaps into focus. A stationary state is one special ket in which every observable's statistics freeze — a property of the state, bought by paying all the probability into a single energy province. A conserved observable freezes its own statistics in every ket — a property of the observable, bought by aligning its provinces with the energy structure. The two intersect (in a stationary state, conserved quantities are doubly still) but neither contains the other: a sloshing double-well packet is violently non-stationary, yet its parity census never moves — the machine run below holds $\langle\hat{\Pi}\rangle$ to $10^{-13}$ while $\langle\hat{x}\rangle$ swings by more than a full length unit.

Figure

Stationary state versus conserved observable — a frozen ket versus a frozen census

The definition and the distribution theorem

Fix a time-independent $\hat{H}$ with evolution $\hat{U}(t)=e^{-i\hat{H}t/\hbar}$ . An observable $\hat{A}$ (no explicit time dependence) is a constant of the motion iff

[\hat{A},\hat{H}]=0 \qquad (1)

where: the average-level consequence is immediate — the Ehrenfest engine (previous article, cited not re-derived) gives $\frac{d}{dt}\langle\hat{A}\rangle=\frac{1}{i\hbar}\langle[\hat{A},\hat{H}]\rangle=0$ . But averages are a weak shadow of what (1) actually buys. The strong statement runs through the evolution operator itself: since $\hat{U}$ is a power series in $\hat{H}$ , anything commuting with $\hat{H}$ commutes with $\hat{U}$ ,

[\hat{A},\hat{H}]=0\ \Longrightarrow\ [\hat{A},\hat{U}(t)]=0\quad\text{and}\quad[\hat{P}_n,\hat{U}(t)]=0 \qquad (2)

where: $\hat{P}_n$ is the eigenprojector onto the (possibly degenerate) eigenspace of eigenvalue $a_n$ — projectors are limits of polynomials in $\hat{A}$ (spectral theory per Projection Operators), so they inherit the commutation. Equation (2) is the border-patrol statement: the provinces of $\hat{A}$ are invariant subspaces of the dynamics. From it, the full conservation theorem falls out in two lines (Derivation A):

P(a_n,t)=\langle\psi(t)|\hat{P}_n|\psi(t)\rangle=P(a_n,0)\qquad\text{for every }n,\ \text{every }|\psi(0)\rangle,\ \text{every }t \qquad (3)

where: all moments follow ( $\langle\hat{A}^k\rangle=\sum_n a_n^kP(a_n)$ — the spectral bookkeeping of Expectation Values and Variance), so the mean, the variance, the whole histogram of a conserved quantity is dynamical bedrock. The verifier exhibits (3) on a random six-level system with a degenerate conserved observable — probabilities frozen to $10^{-13}$ — while a deliberately non-commuting observable's histogram visibly drifts in the same evolution.

Derivation AThe frozen-distribution theorem

We prove that $[\hat{A},\hat{H}]=0$ freezes every outcome probability of $\hat{A}$ , in every state. The strategy is to move the conservation from the observable to its projectors and then let unitarity cancel. Two inputs are consumed with citation: the exponential-series argument that functions of $\hat{H}$ commute with whatever $\hat{H}$ commutes with, and the spectral fact that eigenprojectors are (limits of) polynomials in the observable. The degenerate case costs nothing extra — that is the point of working with projectors rather than eigenvectors.

Step 1. From $[\hat{A},\hat{H}]=0$ to $[\hat{A},\hat{U}]=0$ : expand $\hat{U}(t)=\sum_k\frac{1}{k!}\left(\frac{-it}{\hbar}\right)^k\hat{H}^k$ and commute $\hat{A}$ through term by term ( $[\hat{A},\hat{H}^k]=0$ by induction from (1)).

Step 2. From $\hat{A}$ to its projectors: each $\hat{P}_n$ is a polynomial in $\hat{A}$ (e.g. $\hat{P}_n=\prod_{m\neq n}\frac{\hat{A}-a_m}{a_n-a_m}$ for finite spectra), so $[\hat{P}_n,\hat{U}]=0$ . Why projectors and not eigenkets: in a degenerate eigenspace individual eigenvectors can rotate under the dynamics — only the eigenspace as a whole is invariant. The projector sees exactly the eigenspace and nothing finer, which is why the theorem survives degeneracy untouched.

Step 3. Freeze the probability. For any initial state, with $|\psi(t)\rangle=\hat{U}|\psi(0)\rangle$ and the degenerate-safe Born rule $P(a_n)=\langle\psi|\hat{P}_n|\psi\rangle$ (Born Rule and Measurement Probabilities):

P(a_n,t)=\langle\psi(0)|\hat{U}^\dagger\hat{P}_n\hat{U}|\psi(0)\rangle=\langle\psi(0)|\hat{P}_n\,\hat{U}^\dagger\hat{U}|\psi(0)\rangle=P(a_n,0)

Why this is strictly stronger than the Ehrenfest one-liner: frozen $\langle\hat{A}\rangle$ constrains one number; frozen $\{P(a_n)\}$ constrains the entire distribution — variance, tails, everything. The converse fails: a cleverly tuned non-conserved observable can have momentarily constant mean while its histogram churns.

Step 4. Good quantum numbers. If $|\psi(0)\rangle$ lies entirely in eigenspace $n$ ( $\hat{P}_n|\psi(0)\rangle=|\psi(0)\rangle$ ), then $\hat{P}_n|\psi(t)\rangle=\hat{U}\hat{P}_n|\psi(0)\rangle=|\psi(t)\rangle$ — the state never leaves the province, and the label $a_n$ is trustworthy bookkeeping for all time. Verified: probabilities frozen to $10^{-13}$ (all outcomes, degenerate level included) and moments $\langle\hat{A}\rangle,\langle\hat{A}^2\rangle,\langle\hat{A}^3\rangle$ frozen to $10^{-12}$ on a random 6-level system, with a non-commuting contrast observable's distribution moving by $>10^{-2}$ in the same run; $[\hat{P}_n,\hat{U}]=0$ checked projector by projector (article verifier).

The algebra of constants

Constants of the motion do not merely accumulate — they generate each other. The classical seed is Poisson's 1809 observation, proved cleanly by Jacobi: the Poisson bracket of two integrals of motion is again an integral. Quantization keeps the theorem verbatim, with the commutator as the bracket:

[\hat{A},\hat{H}]=0,\ [\hat{B},\hat{H}]=0\ \Longrightarrow\ \big[\tfrac{1}{i}[\hat{A},\hat{B}],\hat{H}\big]=0 \qquad (4)

where: the factor $1/i$ keeps the new constant hermitian ( $i[\hat{A},\hat{B}]$ is hermitian for hermitian $\hat{A},\hat{B}$ — the algebra pack of The Commutator and Compatible Observables).

Go deeper · Derivation B · 1 minPoisson's theorem quantized — the harvest of constants

Derivation BPoisson's theorem quantized — the harvest of constants

We prove equation (4): the commutator of two constants of the motion is itself a constant. The strategy is to spend the two given zeros inside the Jacobi identity — the cyclic identity Jacobi supplied to complete Poisson's classical claim, proven for commutators in The Commutator and Compatible Observables and consumed here by citation. We then take stock of what the harvest can and cannot yield, because the termination of the process is as informative as the process itself.

Step 1. Write the Jacobi identity for the triple $(\hat{A},\hat{B},\hat{H})$ :

\big[[\hat{A},\hat{B}],\hat{H}\big]+\big[[\hat{B},\hat{H}],\hat{A}\big]+\big[[\hat{H},\hat{A}],\hat{B}\big]=0

Step 2. Insert the hypotheses $[\hat{B},\hat{H}]=0$ and $[\hat{H},\hat{A}]=0$ : the last two triples vanish identically, leaving

\big[[\hat{A},\hat{B}],\hat{H}\big]=0

Why hermiticity needs the $1/i$ : for hermitian $\hat{A},\hat{B}$ the raw commutator is anti-hermitian; $\frac{1}{i}[\hat{A},\hat{B}]$ is hermitian and commutes with $\hat{H}$ — a legitimate new observable, conserved.

Step 3. Termination honesty. The harvest cannot run forever: a $d$ -level system has only $d^2$ independent operators, so repeated bracketing must eventually re-express earlier constants — the set of constants closes into an algebra. If $\hat{H}$ is nondegenerate the closure is trivial (every constant is a function of $\hat{H}$ , all mutually commuting); degeneracy is the only fertile ground, since only inside a degenerate eigenspace can conserved block operators fail to commute with each other. Why the closed algebra matters: a non-abelian closure — like the angular-momentum triple $[\hat{L}_x,\hat{L}_y]=i\hbar\hat{L}_z$ , next chapter's showcase — is the fingerprint of a symmetry group, and its representation theory will organize entire spectra. The name of that structure, and where the constants come from, is the next article's theorem. *Verified: on a 6-level system with one degenerate pair, two conserved block operators bred a genuinely new conserved constant ( $\|[\hat{A}_3,\hat{B}_3]\|>0.1$ , $[[\hat{A}_3,\hat{B}_3],\hat{H}]=0$ to $10^{-13}$ ); the Jacobi identity expanded symbolically to zero (article verifier).*

Two consequences deserve names. First, the harvest: from a few known constants, commutation manufactures more. Second, the termination honesty of Step 3: what the closure looks like — abelian labeling versus non-abelian symmetry algebra — is a structural fingerprint of the system. The verifier stages a minimal non-trivial harvest: on a six-level system with one degenerate energy pair, two independently conserved block operators $\hat{A}_3,\hat{B}_3$ (σ-like inside the degenerate province) produce a genuinely new conserved $\frac{1}{i}[\hat{A}_3,\hat{B}_3]$ — checked $\neq0$ and commuting with $\hat{H}$ to $10^{-13}$ . Degeneracy is exactly where the harvest is fertile, a hint the next article turns into a theorem. One more structural lemma completes the toolkit — conserved quantities refuse to mix:

[\hat{A},\hat{H}]=0\ \Longrightarrow\ \langle a'|\hat{H}|a\rangle=0\quad\text{whenever }a'\neq a \qquad (5)

where: sandwich $0=\langle a'|[\hat{A},\hat{H}]|a\rangle=(a'-a)\langle a'|\hat{H}|a\rangle$ and divide by $a'-a\neq0$ . The Hamiltonian is block-diagonal in the provinces of every conserved observable — the structural reason good quantum numbers make Hamiltonians tractable (diagonalize block by block; the parity-sorted well spectra of Parity were this lemma in action) and the germ of selection rules: a perturbation commuting with $\hat{A}$ inherits the same blocks and cannot drive $a\to a'$ transitions. The full selection-rule industry (dipole rules, angular momentum) belongs to the angular-momentum and time-dependent chapters. Finally, constants pay a labeling dividend. By the compatibility theorem (The Commutator and Compatible Observables), (1) grants $\hat{A}$ and $\hat{H}$ a joint eigenbasis; adding further mutually-commuting constants extends the joint labels until states are uniquely addressed — the CSCO construction (Simultaneous Diagonalization and Complete Sets of Commuting Observables), now with the dynamical guarantee that every label in the address survives evolution. That is why the hydrogen state $|n,\ell,m\rangle$ is a usable address: each slot is a constant of the motion's eigenvalue.

Limiting cases

case	result	meaning
$\hat{A}=\hat{H}$	energy conserved	$[\hat{H},\hat{H}]=0$ — the first citizen; distribution of energy frozen
$\hat{A}=\hat{1}$	norm conserved	the trivial constant; unitarity's shadow
free particle, $\hat{A}=\hat{p}$	momentum census frozen	$[\hat{p},\hat{p}^2/2m]=0$ ; plane-wave weights never redistribute
symmetric $V$ , $\hat{A}=\hat{\Pi}$	parity frozen	$[\hat{H},\hat{\Pi}]=0$ (Parity); the double-well demo below
$\partial_t\hat{A}\neq0$	not a constant even if $[\hat{A},\hat{H}]=0$	the definition has two clauses; a lab-ramped observable can defeat the commutator
$[\hat{A},\hat{H}]\neq0$ but $\langle[\hat{A},\hat{H}]\rangle=0$ in $\|\psi\rangle$	$\langle\hat{A}\rangle$ momentarily flat	not conservation — higher moments and other states still move

Common trapA conserved observable does not mean nothing about it ever changes in the lab

Three distinct disappointments hide behind the theorem, and each has a different cause. (i) Single-shot outcomes of a conserved $\hat{A}$ are still random — conservation freezes the distribution, not the dice; only a state already inside one province gives certainty. (ii) Measurement is not evolution: a measurement of a non-commuting observable collapses the state (Measurement and Wavefunction Collapse) and re-deals the $\hat{A}$ census — conservation holds between measurements, under $\hat{U}$ alone. (iii) Conservation is model-relative: parity is conserved by the symmetric double well, and broken the moment a real apparatus tilts the potential by a microvolt. When an experiment shows a "conserved" quantity drifting, the commutator is indicting your Hamiltonian, not the theorem.

Worked examples

Example 1 — parity as border patrol in a double well (machine run). A Gaussian packet ( $\sigma^2=\tfrac12$ ) starts at $x_0=1$ , off-center inside the right well of the symmetric potential $V=(x^2-4)^2/8$ (dimensionless units). Track $\langle\hat{\Pi}\rangle$ and $\langle\hat{x}\rangle$ .

Step 1. Check the criterion: $V(-x)=V(x)$ , so $[\hat{H},\hat{\Pi}]=0$ (Parity's theorem) — parity is a constant of the motion. Position is not: $[\hat{x},\hat{H}]=i\hbar\hat{p}/m\neq0$ .

Step 2. Initial census: the displaced Gaussian is neither even nor odd; its parity expectation computes to $\langle\hat{\Pi}\rangle_0=0.135$ — a 57%/43% split between even and odd provinces (from $P_\pm=(1\pm0.135)/2$ ).

Step 3. Evolve on a grid (split-step, 4000 steps): the packet sloshes in the right well, $\langle\hat{x}\rangle$ ranging over $[0.43,1.63]$ — visibly violent, fully non-stationary dynamics.

Step 4. The census: $\langle\hat{\Pi}\rangle$ holds its initial value with maximum drift $7\times10^{-14}$ — machine precision — across the entire run. Parity frozen; position sloshing. The packet will eventually tunnel to the left well and back, $\langle\hat{x}\rangle$ excursing over both wells; the even/odd split of 57/43 will still be exact. A stationary state would have frozen everything; a conserved observable froze exactly its own books.

Example 2 — frozen energy census in a sloshing well. An electron in an ISW of width $L=2$ nm starts in $(\psi_1+\psi_2)/\sqrt{2}$ . What moves, and what cannot?

Step 1. Ground level: $E_1=\pi^2\hbar^2/2m_eL^2=$ 0.094 eV (a quarter of the 1 nm value, by the $1/L^2$ scaling of Infinite Square Well).

Step 2. What moves: the position census. The two components beat at the Bohr frequency, sweeping the centroid sinusoidally with period $T=h/(E_2-E_1)=h/3E_1=$ 14.7 fs and amplitude $16L/9\pi^2=$ 0.360 nm (formula owned by Expectation Values and Variance's wave-register twin, applied here at fresh width).

Step 3. What cannot move: the energy census. $\hat{H}$ is conserved trivially, so $P(E_1)=P(E_2)=\tfrac12$ — exactly, forever (machine-checked to $10^{-15}$ ). No amount of sloshing redistributes energy weight; a time-independent Hamiltonian never causes transitions (the frozen-populations doctrine of Spectral Decomposition of Time Evolution, now recognized as this article's theorem applied to $\hat{A}=\hat{H}$ ).

Step 4. The exam-grade summary: to change an energy census you must break clause (1) — couple a time-dependent term into $\hat{H}$ . That is precisely the business of time-dependent perturbation theory, four chapters on.

Origins · 1 minCarl Gustav Jacob Jacobi and how this was discovered

Carl Gustav Jacob Jacobi

1804–1851 · Königsberg, Berlin

The theorem that constants of the motion breed — equation (4) — carries Poisson's name but earned its proof from Jacobi, who found the gap and filled it with the identity that now underwrites the quantum version. Poisson had announced in 1809 that the bracket of two integrals of motion is an integral; the argument was incomplete, and for three decades the claim sat as celestial-mechanics folklore. Jacobi, the Königsberg virtuoso whose Vorlesungen über Dynamik (1842–43) turned mechanics into systematic bracket calculus, saw that the theorem is a one-line corollary of a deeper algebraic fact — the cyclic identity $\{\{A,B\},C\}+\{\{B,C\},A\}+\{\{C,A\},B\}=0$ — which he proved and which physics now calls the Jacobi identity. His harvest doctrine was explicitly practical: given two integrals of a mechanical problem, bracket them and either reap a new integral or learn that the algebra has closed, and his lectures ran the method on tops, geodesics and planetary motion. Quantum mechanics inherited the entire apparatus unmodified — the commutator satisfies Jacobi's identity, so his closure theorem holds with hats on, and the closed algebras his method terminates in are today's symmetry groups. He worked himself into breakdowns, feuded generously, and died at 47 of smallpox; the identity is on duty in every Lie algebra computation physics performs. (Poisson's own portrait, and the bracket's birth, live with the Heisenberg equation of motion; the abstract bracket algebra with Sophus Lie at the commutator article.)

History

1809

— Poisson announces: the bracket of two integrals of motion is an integral (proof incomplete)

1842

— Jacobi's Königsberg lectures supply the identity and the clean proof; the harvest method is born

1918

— Noether ties conservation to symmetry in full generality for classical field theory

1925

— Dirac's correspondence: commutators inherit the bracket algebra, Jacobi identity included

1926

— Born–Heisenberg–Jordan use commuting-with-H operators to label matrix-mechanics spectra

1929

— Weyl's group-theoretic quantum mechanics makes good quantum numbers systematic

Going further

References. Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §3.5 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §2.1 & §4.1 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Ch. III · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 11 · Jacobi, Vorlesungen über Dynamik (1842–43; Reimer 1866).

Provenance. All equations are standard textbook results. A verifier should recompute: (1) the Ehrenfest derivation $\frac{\mathrm{d}}{\mathrm{d}t}\langle A\rangle=\frac{i}{\hbar}\langle[H,A]\rangle+\langle\partial_t A\rangle$ from the Schrödinger equation (Griffiths §3.5, Sakurai §2.1); (2) $[A,H]=0\Rightarrow[A,U(t)]=0$ via the exponential series (Sakurai §2.1); (3) the frozen-distribution result using shared eigenbasis (Cohen-Tannoudji Ch. III); (4) the symmetry-generator commutator from $U(s)HU^\dagger(s)=H$ (Sakurai §4.1, Shankar Ch. 11); (5) the momentum example $[p,V]=-i\hbar\,\partial_x V$ from $[x,p]=i\hbar$ (Griffiths §2.4, §3.5). No numerical constants beyond $\hbar$ are asserted. For this edition items (2)–(3) were machine-executed (numpy/sympy): frozen probabilities/moments to $10^{-13}$ on a degenerate 6-level system with moving non-commuting contrast; $[\hat{P}_n,\hat{U}]=0$ projector-wise; the Jacobi closure with a genuinely new constant harvested inside a degenerate block; the no-mixing block structure; the double-well parity run ( $\langle\Pi\rangle$ drift $7\times10^{-14}$ , $\langle x\rangle$ range 1.2); and the ISW numbers ( $E_1=0.094$ eV at 2 nm, beat 14.7 fs, centroid amplitude 0.360 nm). Item (1) is owned by Ehrenfest's Theorem (cited); item (4) — the symmetry origin of constants — is deliberately NOT executed here: it is the next article's theorem (Symmetries and Conservation in Quantum Dynamics). Per the chapter ownership map: projector theory is Projection Operators'; the degenerate Born rule is Born Rule and Measurement Probabilities'; compatibility/CSCO labeling is consumed from The Commutator and Compatible Observables and Simultaneous Diagonalization; the stationary-state contrast cashes Stationary States' explicit deferral; Zeno-type measurement dynamics → Quantum Zeno Effect by name.

Builds on: Ehrenfest's Theorem · The Heisenberg Equation of Motion · The Commutator and Compatible Observables · Projection Operators · Stationary States and Energy Eigenstates

Leads to: Symmetries and Conservation in Quantum Dynamics · Simultaneous Diagonalization and Complete Sets of Commuting Observables · Angular momentum (the harvest's showcase) · Quantum Zeno Effect

Is a constant of the motion the same as a stationary state?

No — they are dual freezings, and the article's second figure is the honest summary. A stationary state is one ket in which all observables' statistics freeze; a conserved observable is one quantity whose statistics freeze in all kets. Neither implies the other: a sloshing double-well packet is maximally non-stationary yet keeps its parity census exactly; conversely, in a stationary state even non-conserved observables like $\hat{x}$ have frozen statistics. The stationary-states article deferred this distinction here by name; consider it cashed.

If individual eigenvectors can rotate inside a degenerate eigenspace, in what sense is the label conserved?

The label names the province, not a particular ket inside it. A state starting in the $a_n$ eigenspace stays in it ( $\hat{P}_n$ invariance, Step 4), so measurements of $\hat{A}$ return $a_n$ with certainty forever — that is what the quantum number guarantees. What may evolve is which vector within the province the state is — dynamics can be rich inside a border. To pin the state uniquely you add more commuting constants until the CSCO address is complete; each added label is another frozen census.

Does the theorem survive when the Hamiltonian depends on time?

The clean chain breaks at its first link: $\hat{U}$ is then a time-ordered exponential, and $[\hat{A},\hat{H}(t)]=0$ at each instant no longer guarantees $[\hat{A},\hat{U}]=0$ unless $\hat{A}$ commutes with $\hat{H}(t)$ at every time — in which case everything goes through. A weaker tool survives regardless: the Ehrenfest engine with $\hat{H}(t)$ inserted. And one genuinely new possibility opens: a quantity conserved by $\hat{H}_0$ can be pumped by a drive — exactly how time-dependent perturbation theory moves population between provinces the static theory declared sealed.

Why does the article call energy "the first citizen" — isn't energy conservation trivial?

It is the one conservation law that costs nothing: $[\hat{H},\hat{H}]=0$ identically, so a static Hamiltonian conserves its own distribution in every state — no symmetry analysis required. Non-trivial is everything downstream: which other operators join $\hat{H}$ (that is system-specific), and the fact that the frozen energy census is much stronger than frozen $\langle H\rangle$ — populations $|c_n|^2$ never redistribute, which is why a time-independent Hamiltonian cannot cause transitions no matter how long you wait. Trivial criterion, powerful theorem.

Can the Jacobi harvest produce infinitely many constants?

Not independently. In a $d$ -dimensional Hilbert space there are only $d^2$ linearly independent operators, so brackets must eventually re-express earlier constants — the algebra closes. What the closure looks like is deeply informative: an abelian closure (everything commutes) gives a CSCO-style labeling; a non-abelian closure (like the angular-momentum triple, coming next chapter) forces degeneracy and signals a symmetry group. Classically the analogous count caps independent integrals at $2n-1$ for $n$ degrees of freedom, with "maximally closing" systems being the integrable ones — the quantum echo of that story is the integrability/thermalization frontier mentioned above.

A quantity is conserved in my model but drifts in my experiment. Which theorem failed?

None — your Hamiltonian did. Conservation is always relative to a model $\hat{H}$ : parity holds until a stray field tilts the well; $\hat{S}_z$ holds until a transverse fluctuation appears. Experimentally this is a feature: measuring the drift rate of a nominally conserved quantity is a precision probe of the neglected terms (that is how atomic clocks bound exotic symmetry violations). The debugging order is: (i) check $\partial_t\hat{A}$ — is the observable itself lab-ramped? (ii) recompute $[\hat{A},\hat{H}_{\rm full}]$ with all couplings; (iii) remember measurements themselves re-deal the census (the warning box's clause).