Standard Deviation and Quantum Uncertainty

Expectation Values taught you the mean of a quantum measurement's outcome distribution. This article measures what the mean hides: the spread. Two states can share the same $\langle x\rangle$ while one nails the particle to a nanometre and the other smears it across a micron — and everything distinctive about quantum mechanics lives in that difference. The right ruler is the classical statistician's standard deviation,

\sigma_Q = \sqrt{\langle Q^2\rangle - \langle Q\rangle^2},

the root-mean-square distance of outcomes from their mean. Nothing quantum yet: dice have a $\sigma$ , and computing it is a warm-up here. The quantum content arrives in three installments. First, $\sigma$ is a preparation property, not an instrument error — a perfect detector still records scatter, because the state itself carries a distribution. Second, $\sigma_Q = 0$ is possible and special: it happens exactly in the eigenstates of $\hat Q$ , closing the certainty loop that Operators and Observables opened. Third — the headline — the spreads of position and momentum in the same state are chained together:

\sigma_x\,\sigma_p \geq \frac{\hbar}{2},

the Heisenberg uncertainty principle. No state, however cleverly prepared, squeezes both below the floor. This article builds the $\sigma$ craft honestly — oscillator and box states computed to the digit — states the principle with its wave-mechanical plausibility, and hands the formal proof to its rightful owner in the Formalism chapter. What you own afterwards is the fluctuation grammar of the entire track: every " $\Delta E$ ", every linewidth, every squeezing claim in a laser paper is this article's $\sigma$ .

Key takeaways

The spread of an observable's outcome distribution: $\sigma_Q^2 = \langle(Q - \langle Q\rangle)^2\rangle = \langle Q^2\rangle - \langle Q\rangle^2$ — the second form is the workhorse.
$\sigma$ is a property of the state (the preparation), not of the apparatus: quantum scatter survives perfect instruments.
$\sigma_Q = 0 \iff$ the state is an eigenstate of $\hat Q$ — the converse of the certainty theorem, proved here.
Computing craft: get $\langle Q\rangle$ and $\langle Q^2\rangle$ by sandwich or eigenvalue bookkeeping; subtract; root.
QHO ground state: $\sigma_x\sigma_p = \hbar/2$ exactly — the minimum-uncertainty state, integrals done in full here.
Box states: $\sigma_x\sigma_p = 0.568\hbar$ ( $n=1$ ), $1.67\hbar$ ( $n=2$ ) — always above the floor, growing up the ladder.
Key idea: standard deviation quantifies the intrinsic indeterminacy a quantum state assigns to an observable — and for position and momentum the two indeterminacies can never both be small: $\sigma_x\sigma_p \geq \hbar/2$ .

The core insight

The standard deviation $\sigma_Q = \sqrt{\langle Q^2\rangle - \langle Q\rangle^2}$ measures the spread of measurement outcomes about the mean — an intrinsic property of the prepared state, vanishing exactly on eigenstates, and bounded for conjugate pairs by $\sigma_x\sigma_p \geq \hbar/2$ .

Where this lives today

The uncertainty floor is now an engineering budget line. LIGO's detectors are limited across their frequency band by quantum noise traceable to vacuum fluctuations — position-like and momentum-like quadratures of light obeying exactly a $\sigma_A\sigma_B$ floor — and since 2019 both LIGO and Virgo inject squeezed vacuum: light engineered with $\sigma$ below the vacuum level in the quadrature being read, the excess pushed into the conjugate one, precisely the along-the-floor slide of this article's second figure. The 2023-era frequency-dependent squeezing upgrade rotates the squeezed quadrature with frequency, buying a broadband sensitivity gain that has directly increased the black-hole-merger catalogue. Atomic clocks and quantum sensors run the same ledger with atoms: $N$ uncorrelated atoms give clock instability scaling as the "standard quantum limit" $\sigma \propto 1/\sqrt N$ (Bernoulli statistics on quantum coins), and spin-squeezed ensembles — atoms entangled so the collective spin's $\sigma$ shrinks in the measured direction — now beat that limit in deployed optical-lattice clocks. Every such paper quotes its gain in decibels of variance reduction: this article's $\sigma^2$ , in logarithmic clothing.

The physical picture

Two archers with identical average points of impact are not identical archers: one groups arrows inside a coin, the other scatters them across the target. The mean cannot see the difference; the standard deviation is built to. Take every outcome's distance from the mean, square it (so left and right misses cannot cancel), average, and take the root to restore units. Small $\sigma$ : tight grouping, sharp predictions. Large $\sigma$ : the mean is a poor summary of any single shot. Quantum states are archers. Prepare an ensemble in $\Psi$ and measure position once per copy, as the factory line of Expectation Values prescribes: the histogram has a balance point $\langle x\rangle$ and a width $\sigma_x$ — and the width is not sloppiness. It is written into $|\Psi|^2$ itself, survives ideal instruments, and differs from state to state in physically decisive ways.

Figure

Same mean, different states: the width is physics the mean cannot see

The uncertainty principle is then a statement about pairs of histograms drawn from the same archer. Measure position on half your ensemble and momentum on the other half: the two widths are not independent dials. The corkscrew picture of Momentum Operator says why. A state sharp in momentum is a long, evenly-wound corkscrew — inevitably spread in position. A state sharp in position is a thin spike — a chord containing every winding rate at once, spread in momentum. The Fourier reciprocity between the two languages ( $\Phi$ as the transform of $\Psi$ , previewed there and owned by Momentum-Space Wavefunction) hard-wires the trade-off; $\hbar/2$ is its exact exchange rate.

Figure

The uncertainty ledger: states may move along the bound, never below it

The spread, built properly

Definition first, in the language of Expectation Values — everything is an expectation value of something:

\sigma_Q^2 \equiv \left\langle\left(Q - \langle Q\rangle\right)^2\right\rangle \qquad (1)

where: $Q - \langle Q\rangle$ — the deviation from the mean, itself an honest observable (Hermitian: a Hermitian $\hat Q$ minus a real constant); squaring before averaging prevents the cancellation that makes $\langle Q - \langle Q\rangle\rangle = 0$ useless as a width. The square root of $\sigma_Q^2$ — the variance — restores the observable's units: $\sigma_x$ is a length, $\sigma_p$ a momentum, $\sigma_H$ an energy.

Derivation AThe computational form — and why the two definitions agree

Definition (1) is conceptually right and computationally clumsy: it requires knowing $\langle Q\rangle$ before you integrate. The identity $\sigma_Q^2 = \langle Q^2\rangle - \langle Q\rangle^2$ needs two independent moments instead, each a standard sandwich. The proof is three lines of expectation algebra, and the byproducts are permanently useful: linearity of the mean, and the always-true inequality $\langle Q^2\rangle \geq \langle Q\rangle^2$ that already appeared as a warning in Expectation Values.

Step 1. Expand the square. $\left(Q - \langle Q\rangle\right)^2 = Q^2 - 2\langle Q\rangle\,Q + \langle Q\rangle^2$ . Why: $\langle Q\rangle$ is a number, not an operator — it commutes with everything and squares like a number.

Step 2. Average term by term. $\sigma_Q^2 = \langle Q^2\rangle - 2\langle Q\rangle\langle Q\rangle + \langle Q\rangle^2$ , using linearity of the expectation (the integral in the sandwich is linear) and $\langle c\rangle = c$ for constants. Why: the middle term's inner $\langle Q\rangle$ rides through the outer average untouched.

Step 3. Collect.

\sigma_Q^2 = \langle Q^2\rangle - \langle Q\rangle^2 \qquad (2)

Why: two sandwiches — one with $\hat Q$ , one with $\hat Q^2$ — and a subtraction. Since $\sigma_Q^2 \geq 0$ by construction (it averages a square), the inequality $\langle Q^2\rangle \geq \langle Q\rangle^2$ rides along free, with equality exactly when the distribution has zero width. *Verified: sympy confirms the identity symbolically and on the die distribution below; standard statistics, Griffiths §1.5.*

Go deeper · Derivation B · 1 minZero spread forces an eigenstate — the certainty loop closed

Derivation BZero spread forces an eigenstate — the certainty loop closed

Operators and Observables proved eigenstate ⇒ every measurement returns the eigenvalue. Here is the converse, which it explicitly deferred: if a state's spread in $Q$ vanishes, the state must be an eigenfunction of $\hat Q$ . Together the pair says: certainty states and eigenstates are the same set, exactly. The proof is a lovely use of Hermiticity — the variance becomes the norm of a vector, and only the zero vector has zero norm.

Step 1. Write the variance as a sandwich of a square. $\sigma_Q^2 = \int\Psi^*\,(\hat Q - q)^2\,\Psi\,dx$ with $q \equiv \langle Q\rangle$ . Why: definition (1), with the deviation operator $\hat A \equiv \hat Q - q$ — Hermitian, since $q$ is real.

Step 2. Split the square using Hermiticity. $\int\Psi^*\hat A(\hat A\Psi)\,dx = \int(\hat A\Psi)^*(\hat A\Psi)\,dx = \int|\hat A\Psi|^2dx$ . Why: move one factor of $\hat A$ leftward across the sandwich — the licensed move of Operators and Observables — and the variance is revealed as the squared length of the function $\hat A\Psi$ .

Step 3. Demand zero. $\sigma_Q = 0 \Rightarrow \int|\hat A\Psi|^2dx = 0 \Rightarrow \hat A\Psi = 0$ everywhere (a non-negative integrand with zero integral vanishes identically). Why: no cancellation is possible inside $|\cdot|^2$ — this is where "squared" earns its keep.

Step 4. Read the eigenvalue equation. $\hat A\Psi = 0$ is $\hat Q\Psi = q\Psi$ : the state is an eigenfunction, with its mean as the eigenvalue. Why: the loop closes — $\sigma_Q = 0 \iff \hat Q\Psi = q\Psi$ . Determinate states are eigenstates, no exceptions; and since $\hat x$ and $\hat p$ share no eigenfunctions (a delta spike is no corkscrew), no state can have $\sigma_x = \sigma_p = 0$ — the uncertainty principle's qualitative core, before any inequality. *Verified: sympy checks Step 2's Hermiticity split on concrete states and the ISW eigenstate instance; Griffiths §3.2.*

With the craft in place, the principle itself. For position and momentum in any normalizable state,

\sigma_x\,\sigma_p \geq \frac{\hbar}{2} \qquad (3)

where: both spreads are computed in the same state, via (2). The wave-mechanical reading was the second figure's: sharp momentum needs many coherent turns of the corkscrew (long state), sharp position needs a short state — and the Fourier bandwidth theorem makes "you cannot have both" quantitative, with Gaussians as the unique states that saturate. What this article does not do is prove (3) in generality: the clean proof runs through the Schwarz inequality and the commutator $[\hat x, \hat p] = i\hbar$ of Momentum Operator, delivers the general form $\sigma_A\sigma_B \geq \tfrac12|\langle[\hat A,\hat B]\rangle|$ for any pair of observables, and is owned by the Formalism chapter's Generalized Uncertainty Principle. What this article does instead is what the proof cannot: make the numbers concrete — a state that sits exactly on the floor, and a family that sits provably above it.

Limiting cases

Case	Result	Meaning
eigenstate of $\hat Q$	$\sigma_Q = 0$	certainty; the converse theorem makes the set exact
classical die	$\sigma = 1.71$ about mean $3.5$	the machinery is ordinary statistics — nothing quantum yet
QHO ground state	$\sigma_x\sigma_p = \hbar/2$ exactly	the unique-shape floor: Gaussian, worked below
ISW $n = 1$	$0.568\,\hbar$	above the floor — hard walls forbid Gaussian tails
ISW $n \gg 1$	$\sigma_x\sigma_p \approx n\pi\hbar/\sqrt{12}$ , growing	excited states drift far from minimum uncertainty
$\sigma_x \to 0$ (position spike)	$\sigma_p \to \infty$	the delta-state pathology of Position Operator, quantified

Common trap“The uncertainty principle says measurement disturbs the particle”

That is a different (and historically earlier) idea, and conflating the two costs precision. Relation (3) is a statement about preparations: one ensemble, two histograms — position spreads measured on some copies, momentum spreads on others, no copy measured twice, nothing disturbed. The widths fail to both be small because the state itself cannot encode both sharpnesses, exactly as a chord cannot have both a precise onset time and a precise pitch. Heisenberg's 1927 microscope story — photons kicking the electron during measurement — is about measurement back-action, a real but distinct phenomenon with its own modern formulation (error–disturbance relations, which are quantitatively different inequalities). Keep the ledgers separate: $\sigma_x\sigma_p \geq \hbar/2$ would hold in a universe of perfectly gentle measurements, because it is a fact about wavefunctions, not about clumsy apparatus. A second trap rides with the first: $\sigma$ is not a resolution limit — single position measurements can be arbitrarily precise; it is the scatter across the ensemble that no preparation can eliminate.

Worked examples

1 · From dice to the oscillator floor (confidence builder)

(a) Warm up classically: compute $\sigma$ for one fair die. (b) Then the quantum flagship: for the QHO ground state $\psi_0 = (m\omega/\pi\hbar)^{1/4}e^{-m\omega x^2/2\hbar}$ (from Harmonic Oscillator, Analytic Method), compute $\sigma_x$ , $\sigma_p$ , and their product.

Step 1. The die: $\langle Q\rangle = \frac{1+2+\cdots+6}{6} = 3.5$ ; $\langle Q^2\rangle = \frac{1+4+\cdots+36}{6} = \frac{91}{6}$ . By (2): $\sigma^2 = \frac{91}{6} - 12.25 = \frac{35}{12}$ , so $\sigma = \boxed{1.71}$ . Note the mean is no possible outcome and $\sigma$ spans the gap to the nearest faces — ordinary statistics, calibrating the machinery.

Step 2. Oscillator, position: by symmetry $\langle x\rangle = 0$ ; the Gaussian integral gives $\langle x^2\rangle = \dfrac{\hbar}{2m\omega}$ , so $\sigma_x = \sqrt{\dfrac{\hbar}{2m\omega}}$ .

Step 3. Oscillator, momentum: $\langle p\rangle = 0$ (real state — the rule from Momentum Operator); the sandwich $\langle p^2\rangle = -\hbar^2\int\psi_0\psi_0''\,dx = \dfrac{m\omega\hbar}{2}$ , so $\sigma_p = \sqrt{\dfrac{m\omega\hbar}{2}}$ .

Step 4. The product: $\sigma_x\sigma_p = \sqrt{\dfrac{\hbar}{2m\omega}\cdot\dfrac{m\omega\hbar}{2}} = \boxed{\dfrac{\hbar}{2}}$ — exactly on the floor, with $m$ and $\omega$ cancelling completely. This is the arithmetic behind the saturation claim recalled from Harmonic Oscillator, Analytic Method: the Gaussian is the unique shape that makes (3) an equality, and the oscillator's ground state is that shape. Cross-check: $\langle H\rangle = \dfrac{\sigma_p^2}{2m} + \dfrac{m\omega^2\sigma_x^2}{2} = \dfrac{\hbar\omega}{4} + \dfrac{\hbar\omega}{4} = \dfrac{\hbar\omega}{2}$ ✓ — the zero-point energy, split evenly between kinetic and potential fluctuation.

2 · Climbing the box ladder — and the energy spread of a superposition (exam level)

(a) For the ISW states $\psi_n$ , using $\langle x^2\rangle = L^2\left(\tfrac13 - \tfrac{1}{2n^2\pi^2}\right)$ and $\langle p^2\rangle = (n\pi\hbar/L)^2$ (moments owned by Infinite Square Well), find $\sigma_x\sigma_p$ for $n = 2$ and compare with $n = 1$ . (b) For the equal superposition $\Psi = \frac{1}{\sqrt2}(\psi_1 + \psi_2)$ with $L = 1.0\ \mathrm{nm}$ , find $\sigma_H$ .

Step 1. Box spreads: $\langle x\rangle = L/2$ always, so $\sigma_x^2 = L^2\left(\tfrac{1}{12} - \tfrac{1}{2n^2\pi^2}\right)$ ; and $\langle p\rangle = 0$ , so $\sigma_p = n\pi\hbar/L$ .

Step 2. Products: $n = 1$ : $\sigma_x\sigma_p = \pi\hbar\sqrt{\tfrac{1}{12} - \tfrac{1}{2\pi^2}} = 0.568\hbar$ — the number quoted by Momentum Operator and Infinite Square Well, re-earned. $n = 2$ : $\sigma_x\sigma_p = 2\pi\hbar\sqrt{\tfrac{1}{12} - \tfrac{1}{8\pi^2}} = \boxed{1.67\,\hbar}$ — nearly triple the floor. Each extra node adds momentum content at essentially fixed spatial extent; for large $n$ , $\sigma_x \to L/\sqrt{12}$ (the flat-distribution value) and the product grows as $n\pi\hbar/\sqrt{12} \approx 0.91n\hbar$ . Minimum uncertainty is a ground-state luxury.

Step 3. Energy spread of the mix: outcomes $E_1 = 0.376\ \mathrm{eV}$ and $E_2 = 1.504\ \mathrm{eV}$ , each with probability $\tfrac12$ (per Stationary States and Separation of Variables). Then $\langle H\rangle = 0.940\ \mathrm{eV}$ and $\langle H^2\rangle = \tfrac12(E_1^2 + E_2^2) = 1.202\ \mathrm{eV^2}$ , so $\sigma_H = \sqrt{1.202 - 0.884} = \boxed{0.564\ \mathrm{eV}}$ — exactly $(E_2 - E_1)/2$ , as a two-spike distribution demands (half the gap between the spikes).

Step 4. Read $\sigma_H$ physically: this state's energy is irreducibly indefinite by $\pm0.564\ \mathrm{eV}$ — not through ignorance, but because the preparation contains two sharp energies at once. And the beat period of the same state ( $3.67\ \mathrm{fs}$ , from Stationary States and Separation of Variables) is $2\pi\hbar/(2\sigma_H)$ : states with big energy spread evolve fast, states with $\sigma_H = 0$ (eigenstates) do not evolve at all — the energy–time reciprocity in its honest, derived form.

Origins · 1 minWerner Heisenberg and how this was discovered

Werner Heisenberg

1901–1976 · Göttingen, Copenhagen, Leipzig

By early 1927 Heisenberg's own matrix mechanics — the Helgoland story told with the Ladder Operators article — had a rival in Schrödinger's waves, and its inventor had a problem: what did the theory's refusal to assign simultaneous position and momentum mean? Holed up in Copenhagen, arguing nightly with Bohr until both were raw, the 25-year-old produced the February 1927 paper "On the intuitive content of quantum-theoretical kinematics and mechanics" — the uncertainty paper. Its famous gamma-ray-microscope story (short-wavelength photons locate the electron sharply but kick its momentum unpredictably) carried the physics to the world, and carried a flaw: Bohr, correcting the draft, pointed out the argument needed the wave nature of light through the microscope's aperture, not just photon recoil — disturbance alone was not the principle's foundation. The mathematics settled deeper than the story: within months Kennard proved the exact preparation-level inequality $\sigma_x\sigma_p \geq \hbar/2$ for the standard deviations of this article, and Robertson (1929) generalized it to any observable pair via the commutator. Heisenberg's essay had reached for Anschaulichkeit — "intuitability" — and redefined it: the quantum world is intuitable, but only through distributions and their widths. His microscope survives in textbooks as folklore; the $\sigma$ -inequality it gestured at survives as law.

History

1927

— Heisenberg's uncertainty paper: the microscope argument and the first Δx·Δp estimates.

1927

— Kennard proves the exact standard-deviation form σₓσₚ ≥ ℏ/2 for all states.

1928

— Weyl's group-theory text presents the inequality's clean Schwarz-inequality proof.

1929

— Robertson generalizes to arbitrary pairs: σ_Aσ_B ≥ ½|⟨[Â,B̂]⟩|.

1980

— Caves analyses squeezed light for gravitational-wave detection: engineering below-vacuum σ in one quadrature.

Going further

Texts. Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §1.4–1.6 & §3.5 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 9 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §1.4 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Ch. III · Kennard, Z. Phys. 44, 326 (1927).

Provenance. All equations are standard textbook results. A verifier should recompute: (1) the identity $\sigma_Q^2 = \langle Q^2\rangle - \langle Q\rangle^2$ and the die numbers ( $\sigma = \sqrt{35/12} = 1.71$ ); (2) the zero-variance⇒eigenstate argument via the Hermiticity split (Griffiths §3.2); (3) the QHO ground-state integrals $\langle x^2\rangle = \hbar/2m\omega$ , $\langle p^2\rangle = m\omega\hbar/2$ , product exactly $\hbar/2$ , and $\langle H\rangle = \hbar\omega/2$ ; (4) the box products $0.568\hbar$ ( $n=1$ , moments recalled from Infinite Square Well) and $2\pi\hbar\sqrt{1/12 - 1/8\pi^2} = 1.67\hbar$ ( $n=2$ , computed fresh), large- $n$ growth $\approx 0.91n\hbar$ ; (5) $\sigma_H = (E_2 - E_1)/2 = 0.564\ \mathrm{eV}$ for the equal mix at $L = 1\ \mathrm{nm}$ . The general proof of $\sigma_x\sigma_p \geq \hbar/2$ (Schwarz + commutator) is owned by the Formalism chapter's Generalized Uncertainty Principle; packet-level Fourier reciprocity by Momentum-Space Wavefunction and Wave Packets and Their Construction; the energy–time relation's honest form by the Time-Dependent Phenomena chapter.

Prerequisites. The Wavefunction · Born Statistical Interpretation · Probability Density and Normalization · Wave–Particle Duality · Expectation Values · Operators and Observables. Leads to. Time-Independent Schrödinger Equation · Stationary States and Separation of Variables · de Broglie Matter Waves · the Formalism chapter's Generalized Uncertainty Principle.

Is σ_Q something I could reduce with a better experiment?

Only down to the state's own value, never below it. Real experiments record $\sigma_{\text{observed}}^2 = \sigma_{\text{state}}^2 + \sigma_{\text{instrument}}^2$ (independent scatters add in variance): better apparatus shrinks the second term toward zero, and the remainder — the quantum floor of that preparation — is untouchable by instrumentation. This is routinely demonstrated, not just asserted: cold-atom groups image ensembles with resolution far finer than the cloud's $\sigma_x$ and reproduce the $|\Psi|^2$ width shot after shot. To genuinely reduce $\sigma_Q$ you must change the state — squeeze it — and then the conjugate observable's spread pays, per the floor.

Why the standard deviation specifically? Other width measures exist.

Convention plus theorem-friendliness. Alternatives — full width at half maximum, interquartile range, mean absolute deviation — are legitimate and sometimes experimentally handier (spectroscopists quote FWHM daily; for a Gaussian, $\mathrm{FWHM} = 2.35\sigma$ ). The standard deviation wins in theory for two reasons: it is built from moments, which the sandwich machinery of Expectation Values computes directly; and the uncertainty theorems take their clean form in it — the Kennard–Robertson bound is exactly $\sigma_A\sigma_B \geq \tfrac12|\langle[\hat A,\hat B]\rangle|$ , whereas FWHM versions hold only case by case (a distribution can even have infinite $\sigma$ with finite FWHM — the Lorentzian lineshape, which is why spectroscopists avoid $\sigma$ ). Know which ruler a paper is using before comparing widths.

Does σₓσₚ ≥ ℏ/2 mean I can't know both x and p of an electron?

Refine "know." For a single future measurement: you may predict position sharply (prepare small $\sigma_x$ ) or momentum sharply (small $\sigma_p$ ), and the floor limits the joint sharpness of your predictions — that much is exactly the theorem. But past values are different: from a position measurement now and one earlier, you can retrodict a trajectory's momentum to better than the floor — retrodiction is not preparation, and no theorem forbids it. And a joint simultaneous measurement of both on one particle is its own subtle topic: it can be done approximately (heterodyne detection), at the cost of added noise obeying its own inequality (with $\hbar$ , not $\hbar/2$ ). The clean, always-true statement remains the preparation form: no state offers sharp predictions for both.

Where does ΔE·Δt ≥ ℏ/2 fit? Time isn't an operator in this course.

Correct instinct — time enters the Schrödinger equation as a parameter, not an observable, so there is no $\sigma_t$ built like $\sigma_x$ , and the energy–time relation cannot be a Robertson-type theorem. Its honest versions define $\Delta t$ operationally. The Mandelstam–Tamm form: for any observable $Q$ , the time for $\langle Q\rangle$ to shift by one $\sigma_Q$ obeys $\sigma_H\,\Delta t_Q \geq \hbar/2$ — big energy spread permits fast evolution, and the worked example's $\sigma_H$ versus beat period is a hand-computable instance. Lifetime–linewidth: a state that decays in $\tau$ has an energy distribution of width $\Gamma \approx \hbar/\tau$ — the reciprocity met in Conservation of Normalization's optical potentials. Both are theorems; neither involves a time operator; both arrive properly in the Time-Dependent Phenomena chapter.

The ground state of the box has σₓσₚ = 0.568ℏ. Could some cleverer box state reach 0.5ℏ?

No — and the obstruction is instructive. Saturation of the Kennard bound requires a Gaussian wavefunction (the equality condition of the Schwarz-inequality proof), and no Gaussian satisfies hard-wall boundary conditions: the walls force $\psi = 0$ at two points, and a Gaussian is nowhere zero. Any state in the box therefore sits strictly above the floor; the ground state's $0.568\hbar$ is close because a half-sine is nearly Gaussian across its bulk (their shapes differ mainly in the tails the wall amputates). General principle worth keeping: the uncertainty floor is universal, but which states can approach it depends on the boundary conditions of the problem — confinement geometries each have their own "most classical" state.

Is σ_H = 0.564 eV for the superposition a real, measurable energy fuzziness or an artifact of mixing two sharp states?

Fully real, with a measurable fingerprint: dynamics. The energy histogram itself just shows two sharp spikes — each individual measurement is sharp; the spread is across outcomes, like every $\sigma$ in this article. But $\sigma_H$ is not idle bookkeeping: it sets the state's evolution speed (the Mandelstam–Tamm reciprocity above — this state's density sloshes with period $2\pi\hbar/2\sigma_H$ ), it governs interference visibility in energy-sensitive detection, and — if the state couples to radiation — the emitted spectrum's structure. Contrast genuine ignorance of which eigenstate you have (a classical 50/50 gamble): identical energy histogram, zero sloshing, no interference. Distinguishing those two situations experimentally is precisely what density-operator language handles at the track's later register — and the distinction is the whole difference between superposition and ignorance.

Do uncertainties ever multiply differently — why ℏ/2 for x,p but other bounds elsewhere?

The general law makes the pattern transparent: $\sigma_A\sigma_B \geq \tfrac12|\langle[\hat A,\hat B]\rangle|$ (Robertson, proven in the Formalism chapter). The bound is half the magnitude of the commutator's expectation — so $[\hat x,\hat p] = i\hbar$ gives the constant floor $\hbar/2$ , universal because that commutator is a number. Other pairs have state-dependent floors: angular-momentum components obey $\sigma_{L_x}\sigma_{L_y} \geq \tfrac{\hbar}{2}|\langle L_z\rangle|$ , which can vanish in states with $\langle L_z\rangle = 0$ — and commuting observables have no floor at all, which is what "compatible" means. One inequality, the whole taxonomy of quantum trade-offs; position–momentum is simply the pair whose trade is unconditional.