Generalized Uncertainty Principle

The uncertainty principle usually arrives as folklore — "you can't know position and momentum at once" — years before it arrives as a theorem. This article is the theorem. For any two observables $\hat A$ and $\hat B$ , in any state, the product of spreads obeys $\sigma_A\,\sigma_B \ge \tfrac12|\langle[\hat A,\hat B]\rangle|$ : the commutator, the incompatibility fingerprint of two articles ago, reappears as a hard floor under joint sharpness. Set $\hat A = \hat x$ , $\hat B = \hat p$ , quote $[\hat x,\hat p] = i\hbar$ , and Heisenberg's $\sigma_x\sigma_p \ge \hbar/2$ drops out as a corollary — one special case of a bound that equally governs spin components, angular momenta, and quadratures of light. What is remarkable is how little the proof needs. Everything was already on your bench: the variance-as-length identity from Expectation Values and Variance turns each spread into the norm of a vector; the Cauchy–Schwarz inequality — owned and proved by Inner Product and Norm of States, consumed here without re-derivation — bounds their inner product; and the Hermiticity bookkeeping of the commutator article splits that inner product into a real part (a covariance) and an imaginary part (half the commutator). Keep both parts and you get Schrödinger's stronger 1930 inequality; discard the covariance and Robertson's 1929 bound remains. The whole architecture is three moves of geometry. The fine print matters as much as the headline. The bound's right side is a state functional, not a constant — it can vanish for non-commuting observables in special states, which is why "never simultaneously sharp" is reserved for pairs like $\hat x,\hat p$ whose commutator is a number. And the equality case is a solvable condition, whose unique solution for $x$ – $p$ is the Gaussian — the reason minimum-uncertainty packets, oscillator ground states, and coherent states all wear the same bell curve.

Key takeaways

Robertson bound: $\sigma_A\,\sigma_B \ge \tfrac12\,|\langle[\hat A,\hat B]\rangle|$ for any state and any pair of observables — proved from Cauchy–Schwarz plus variance-as-length.
Schrödinger strengthening: $\sigma_A^2\sigma_B^2 \ge \bigl|\tfrac12\langle\{\Delta\hat A,\Delta\hat B\}\rangle\bigr|^2 + \bigl|\tfrac1{2i}\langle[\hat A,\hat B]\rangle\bigr|^2$ — keeps the covariance term Robertson discards.
$x$ – $p$ corollary: $\sigma_x\sigma_p \ge \hbar/2$ , with the CCR quoted from its owner article; state-independent because $[\hat x,\hat p] = i\hbar$ is a number.
The floor is a state functional: $[\hat L_x,\hat L_y] = i\hbar\hat L_z \ne 0$ , yet on an $m = 0$ state the bound reads $\ge 0$ — trivially true, spreads still positive.
Equality iff $(\hat B - \langle\hat B\rangle)|\psi\rangle = i\lambda(\hat A - \langle\hat A\rangle)|\psi\rangle$ with real λ (and zero covariance): for $x$ – $p$ this ODE's unique solution is the Gaussian — $\sigma_x\sigma_p = \hbar/2$ exactly.
Spin saturates too: on $|{+z}\rangle$ , $\sigma_{S_x}\sigma_{S_y} = \hbar^2/4 = \tfrac\hbar2|\langle S_z\rangle|$ — equality with numbers you can check.
Key idea: preparation spreads of two observables cannot both shrink below the commutator's floor — incompatibility, quantified state by state.

The core insight

For Hermitian $\hat A, \hat B$ and any state, $\sigma_A^2\sigma_B^2 \ge \bigl|\tfrac12\langle\{\Delta\hat A,\Delta\hat B\}\rangle\bigr|^2 + \bigl|\tfrac1{2i}\langle[\hat A,\hat B]\rangle\bigr|^2$ — dropping the first term gives Robertson's $\sigma_A\sigma_B \ge \tfrac12|\langle[\hat A,\hat B]\rangle|$ , whose $x$ – $p$ case is Heisenberg's principle.

Where this lives today

The saturation-and-covariance fine print is a working industry. Gravitational-wave interferometers operate against equation (3) in quadrature form: LIGO's arms measure optical phase whose partner quadrature carries the penalty, and injected squeezed vacuum — states engineered to shrink one spread at the other's expense while riding the Schrödinger covariance term — now improves detection range as routine infrastructure. Atomic clocks and magnetometers push the same geometry with spin ensembles (Example 1's floor becomes the "standard quantum limit," beaten by spin squeezing). And the folklore-vs-theorem distinction became experimental physics: neutron-optics and photonics groups have measured error–disturbance tradeoffs violating naive microscope formulas while confirming the modern relations — vindicating precisely the preparation/disturbance separation this article's warning box insists on. When a metrologist says "Heisenberg-limited," the word "Heisenberg" is standing on Robertson's two pages.

The physical picture

Picture the two centered vectors the variance article taught you to build: $|f\rangle = \Delta\hat A|\psi\rangle$ and $|g\rangle = \Delta\hat B|\psi\rangle$ . Their lengths are the spreads — $\|f\| = \sigma_A$ , $\|g\| = \sigma_B$ . Cauchy–Schwarz says two vectors cannot overlap by more than the product of their lengths; the uncertainty principle is that statement run in reverse — the overlap $\langle f|g\rangle$ has a piece guaranteed by the commutator, so the lengths cannot both be small. Incompatible observables push every state in two irreconcilable directions, and geometry collects the toll.

Figure

Uncertainty is Cauchy–Schwarz run backwards — the commutator guarantees an overlap the lengths must pay for

Read the two pieces of the overlap physically. The real part is a covariance — do $A$ -deviations and $B$ -deviations track each other in this state? — a correlation you can often prepare away (choose the state so fluctuations decorrelate), and classical statistics has the identical object. The imaginary part is the quantum novelty: it is set by the operator algebra, identical in every state up to the expectation of the commutator, and no preparation can negotiate with it. Robertson's bound keeps only the non-negotiable piece; Schrödinger's keeps both. That is the entire architecture — the derivation below merely walks it with receipts. One reading to install before the math, because it prevents the classic misinterpretation: everything here concerns preparation spreads — the histogram widths of the variance article, measured on separate ensembles (measure $\hat A$ on ten thousand fresh copies, $\hat B$ on ten thousand more). Nothing in the theorem mentions a measurement of one observable disturbing a subsequent measurement of the other. Disturbance is real physics (the collapse article's Stern–Gerlach chains), and Heisenberg's original microscope heuristic was about it — but the modern theorem quantifies what states are, not what apparatus does. The wave chapter's Standard Deviation and Quantum Uncertainty planted this flag; the proof below finally cashes its deferred promise.

The theorem, in three geometric moves

Derivation ARobertson's bound from Cauchy–Schwarz — and Schrödinger's stronger version for free

We prove $\sigma_A^2\sigma_B^2 \ge \bigl|\tfrac12\langle\{\Delta\hat A,\Delta\hat B\}\rangle\bigr|^2 + \bigl|\tfrac1{2i}\langle[\hat A,\hat B]\rangle\bigr|^2$ (Schrödinger), of which Robertson's $\sigma_A\sigma_B \ge \tfrac12|\langle[\hat A,\hat B]\rangle|$ is the discard-one-term corollary. The strategy: convert spreads to vector lengths (the variance article's norm form), bound their overlap by Cauchy–Schwarz — the inequality is consumed from its owner, Inner Product and Norm of States, not re-derived — and then split the overlap into Hermitian-symmetric and antisymmetric parts, whose reality properties the commutator article certified. Watch that the commutator of centered operators equals that of the originals: subtracting numbers costs nothing.

Step 3. Split the overlap. Using Hermiticity of the centered operators, $\langle f|g\rangle = \langle\psi|\Delta\hat A\,\Delta\hat B|\psi\rangle$ . Decompose the product into its symmetric and antisymmetric halves:

\Delta\hat A\,\Delta\hat B = \frac{\{\Delta\hat A, \Delta\hat B\}}{2} + \frac{[\Delta\hat A, \Delta\hat B]}{2}, \qquad [\Delta\hat A,\Delta\hat B] = [\hat A,\hat B] \qquad (1)

The anticommutator of Hermitians is Hermitian → its expectation is real; the commutator is anti-Hermitian → its expectation is purely imaginary (both facts from the commutator article, machine-checked again here). So the two terms are literally the real and imaginary parts of $\langle f|g\rangle$ , and

|\langle f|g\rangle|^2 = \Bigl(\underbrace{\tfrac12\langle\{\Delta\hat A,\Delta\hat B\}\rangle}_{\text{covariance, real}}\Bigr)^{\!2} + \Bigl(\underbrace{\tfrac{1}{2i}\langle[\hat A,\hat B]\rangle}_{\text{real number}}\Bigr)^{\!2} \qquad (2)

Step 4. Assemble. Chaining Steps 2–3 gives Schrödinger's inequality; a sum of two squares is at least either one, so dropping the covariance yields $\sigma_A^2\sigma_B^2 \ge \tfrac14|\langle[\hat A,\hat B]\rangle|^2$ — take square roots for Robertson. Why one might keep the covariance: for correlated states (squeezed light, tilted Gaussians) the discarded term is large and Robertson is badly loose; Schrödinger's version is the sharp tool, and its covariance term is exactly what "squeezing" reshuffles. *Verified: 300 random Hermitian pairs and states — the chain $\sigma_A^2\sigma_B^2 \ge$ Schrödinger $\ge$ Robertson holds in every trial; commutator expectations purely imaginary and anticommutator expectations real to $10^{-10}$ .*

Specialize once, with the flagship pair. The CCR article (three serials on) owns the derivation of $[\hat x,\hat p] = i\hbar$ ; quoting it, the right side of Robertson's bound is $\tfrac12|\langle i\hbar\rangle| = \hbar/2$ — independent of the state, because the commutator is a multiple of the identity:

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

where: this is the only famous case in which the floor is universal. The wave chapter's ladder of examples — the ISW ground state's $\sigma_x\sigma_p = \hbar\sqrt{\pi^2/12 - 1/2} \approx 0.568\,\hbar$ among them (recomputed symbolically in this article's verifier, ownership of the ladder staying with Standard Deviation and Quantum Uncertainty and the ISW article) — all sit above it, as they must; the Gaussian, next section, sits exactly on it.

Equality: who gets to touch the floor

Go deeper · Derivation B · 1 minThe saturation condition — and why every minimum-uncertainty state is a Gaussian

Derivation BThe saturation condition — and why every minimum-uncertainty state is a Gaussian

We derive the exact condition for equality in Robertson's bound and solve it for the $x$ – $p$ pair. The strategy: saturation must be inherited from the two inequalities chained in the proof — Cauchy–Schwarz (equality iff the vectors are parallel, the condition its owner article proved) and the discarded covariance square (equality iff that term vanishes). Combining the two constraints fixes the proportionality constant to be purely imaginary, and for position–momentum the resulting equation is a first-order ODE with exactly one normalizable solution.

Step 1. Saturate Cauchy–Schwarz: $|g\rangle = c\,|f\rangle$ for some complex $c$ — i.e. $\Delta\hat B|\psi\rangle = c\,\Delta\hat A|\psi\rangle$ .

Step 2. Kill the covariance. With $|g\rangle = c|f\rangle$ , the real part of $\langle f|g\rangle = c\,\sigma_A^2$ must vanish, forcing $c = i\lambda$ with $\lambda$ real:

\Delta\hat B\,|\psi\rangle = i\lambda\,\Delta\hat A\,|\psi\rangle, \qquad \lambda \in \mathbb R \qquad (4)

Why the imaginary constant is the fingerprint: a real $c$ would maximize the covariance and zero the commutator term — saturating the wrong part of the overlap. The $i$ routes all the overlap into the non-negotiable piece.

Step 3. Solve for $x$ – $p$ (centered state, $\langle\hat x\rangle = \langle\hat p\rangle = 0$ ): equation (4) reads $-i\hbar\psi' = i\lambda x\psi$ , a first-order ODE, $\psi' = -(\lambda/\hbar)x\psi$ , with solution $\psi \propto e^{-\lambda x^2/2\hbar}$ — normalizable iff $\lambda > 0$ : the Gaussian, and nothing else. Off-center means just shift and boost: $e^{ip_0x/\hbar}$ times a displaced Gaussian.

Step 4. Confirm the floor is touched at every width. For the Gaussian of width σ: $\sigma_x = \sigma$ and $\sigma_p = \hbar/2\sigma$ (both computed symbolically), so $\sigma_x\sigma_p = \hbar/2$ for all σ — a one-parameter family of minimum-uncertainty packets, trading position sharpness against momentum sharpness along the floor. This is the deep reason the oscillator ground state, the coherent states of the one-dimensional chapter, and every "minimum-uncertainty wave packet" in the literature are Gaussians: they are the unique solutions of (4). Zero-point energy's $\hbar\omega/2$ — derived from this same saturation in the ZPE article — is identical geometry wearing energy units. Verified: symbolically — the Gaussian solves the saturation ODE exactly; its normalization, $\sigma_x^2 = \sigma^2$ , $\sigma_p^2 = \hbar^2/4\sigma^2$ , and the product $\hbar/2$ all exact in sympy.

Figure

The uncertainty floor — a forbidden region, a saturating family, and states that hover above

Spins saturate too, and the numbers are checkable in two lines. On $|{+z}\rangle$ with $\hat A = \hat S_x$ , $\hat B = \hat S_y$ : both spreads are $\hbar/2$ (maximal transverse ignorance), the covariance vanishes, and the floor is $\tfrac12|\langle[\hat S_x,\hat S_y]\rangle| = \tfrac\hbar2|\langle \hat S_z\rangle| = \hbar^2/4$ — met exactly: $\sigma_{S_x}\sigma_{S_y} = \hbar^2/4$ (machine-verified, including the real-λ saturation condition). A spin pointing up is as jointly-sharp in the two transverse components as nature permits, and exactly that sharp.

Limiting cases

case	result	meaning
$[\hat A,\hat B] = 0$	floor $= 0$	compatible pair: joint sharpness available — the commutator article's theorem
$[\hat A,\hat B] = i\hbar$ (number)	$\sigma_x\sigma_p \ge \hbar/2$ in every state	the only state-independent floor; "never both sharp" is legitimate here
operator-valued commutator	floor $= \tfrac12\|\langle[\hat A,\hat B]\rangle\|$ , state by state	$L_x, L_y$ : floor $\propto \|\langle L_z\rangle\|$ , vanishes on $m=0$
eigenstate of $\hat A$	$\sigma_A = 0$ forces floor $= 0$	consistent: eigenstates kill $\langle[\hat A,\hat B]\rangle$ when the bound would otherwise bite
Gaussian packet	$\sigma_x\sigma_p = \hbar/2$ exactly, any width	unique solution of the saturation ODE
correlated (squeezed/tilted) states	Robertson loose; Schrödinger term restores sharpness	covariance is physics, not garnish
$\hbar \to 0$	floor $\to 0$	classical phase space: both coordinates sharp — the correspondence limit

Common trapMeasuring x disturbs p — that is what this inequality says

It is what Heisenberg's 1927 microscope heuristic said, but not what the theorem proves. Robertson's inequality constrains preparation: two histogram widths, measured on separate ensembles of identically prepared systems, with no sequential measurement anywhere in sight. Measurement disturbance is a genuinely different topic — real (the collapse article's Stern–Gerlach chains show it), but quantified by different, subtler inequalities (measurement–disturbance relations, developed since the 2000s, live in the entanglement chapter's generalized-measurement territory along with the experiments that tested them). Keep the two claims apart on exams and in the laboratory: a state with tiny $\sigma_x$ has a huge $\sigma_p$ before anyone measures anything; and a gentle measurement can beat naive disturbance folklore without ever threatening equation (3), which was never about disturbance at all.

Worked examples

Example 1 — spin floor, met exactly (confidence builder). For $|{+z}\rangle$ , verify Robertson's bound for $\hat S_x, \hat S_y$ and check saturation.

Step 1. Spreads: $\langle\hat S_x\rangle = \langle\hat S_y\rangle = 0$ and $\hat S_i^2 = \tfrac{\hbar^2}{4}\mathbb 1$ , so $\sigma_{S_x} = \sigma_{S_y} = \hbar/2$ . Product: $\hbar^2/4$ .

Step 2. Floor: $[\hat S_x,\hat S_y] = i\hbar\hat S_z$ (the spin algebra, quoted from its ANGM owner-to-be), so $\tfrac12|\langle[\hat S_x,\hat S_y]\rangle| = \tfrac\hbar2\,|\langle\hat S_z\rangle| = \tfrac\hbar2\cdot\tfrac\hbar2 = \hbar^2/4$ .

Step 3. Compare: $\boxed{\sigma_{S_x}\sigma_{S_y} = \hbar^2/4 = \text{floor}}$ — exact equality, with the covariance term separately zero and saturation condition (4) holding with $\lambda = 1$ (all machine-verified). A polarized spin is a minimum-uncertainty object.

Step 4. Now tilt the state to the equator ( $|{+x}\rangle$ ): $\langle\hat S_z\rangle = 0$ , the floor collapses to zero, but $\sigma_{S_y} = \hbar/2 > 0$ persists. The bound became trivial; the spreads did not vanish — the floor is a lower bound, never a formula for the spreads themselves.

Example 2 — how Gaussian is "minimum uncertainty"? (exam level). A packet has $\sigma_x = 50\,\mathrm{pm}$ (an electron localized to atomic scale). Find the minimum possible $\sigma_p$ and the corresponding minimum kinetic-energy scale, and identify the state that achieves it.

Step 1. Floor: $\sigma_p \ge \dfrac{\hbar}{2\sigma_x} = \dfrac{1.055\times10^{-34}}{2\times5\times10^{-11}} \approx \boxed{1.05\times10^{-24}\ \mathrm{kg\,m/s}}$ .

Step 2. Achiever: only the Gaussian of width $\sigma_x$ (saturation ODE) — any other 50-pm profile carries strictly more momentum spread.

Step 3. Energy scale: with $\langle\hat p\rangle = 0$ , $\langle\hat T\rangle = \sigma_p^2/2m_e \approx (1.05\times10^{-24})^2/(2\times9.11\times10^{-31}) \approx 6.1\times10^{-19}\,\mathrm J \approx \boxed{3.8\ \mathrm{eV}}$ — atomic-scale confinement costs electron-volt-scale energy, the zero-point logic of the wells chapter recovered from pure geometry.

Step 4. Moral for estimates: $\sigma_p^{\min} = \hbar/2\sigma_x$ is the rigorous version of the folklore $\Delta p \sim \hbar/\Delta x$ ; the factor-of-two discipline matters exactly when an exam asks for "the minimum," and the Gaussian is always the silent witness behind the word.

The relativist who proved the folklore

Origins · 1 minHoward Percy Robertson and how this was discovered

Howard Percy Robertson

1903–1961 · Caltech, Princeton

H. P. Robertson is a rare double citizen of physics history: cosmologists know him for the Robertson–Walker metric — the geometry of every expanding-universe model — while every quantum textbook prints his 1929 Physical Review note, barely two pages, which took Heisenberg's 1927 heuristic and Weyl's Gaussian-case proof and established the general theorem $\sigma_A\sigma_B \ge \tfrac12|\langle[\hat A,\hat B]\rangle|$ for arbitrary observables — the argument this article just ran, Cauchy–Schwarz and all. Kennard had proved the $x$ – $p$ case ( $\hbar/2$ , 1927); Robertson saw the commutator was the whole story; Schrödinger added the covariance term a year later. A Montana-born, Göttingen-trained professor equally fluent in differential geometry and quantum formalism — and, during the war, a chief scientific intelligence officer in London vetting V-weapon assessments — he embodied the era when one person could carry general relativity in one pocket and Hilbert-space geometry in the other. The inequality that made "uncertainty principle" a theorem rather than a slogan is, fittingly, the work of a man remembered mostly for something else entirely.

History

1927

— Heisenberg's microscope heuristic; Kennard proves

\sigma_x\sigma_p \ge \hbar/2

the same year

1928

— Weyl's book presents the Gaussian-case argument

1929

— Robertson: the general commutator bound, in two pages

1930

— Schrödinger adds the anticommutator (covariance) term — the strongest quadratic form

1960

s–80s — coherent and squeezed states turn saturation and covariance into laboratory technology

2003–2013

— measurement–disturbance relations (Ozawa, Busch–Lahti–Werner) separate preparation bounds from disturbance physics, tested with neutrons and photons

Going further

References. Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §3.5 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §1.4 · Shankar, Principles of Quantum Mechanics, 2nd ed., §9.3 · Robertson, Phys. Rev. 34, 163 (1929) · Schrödinger, Sitzungsber. Preuss. Akad. Wiss. (1930) · Kennard, Z. Phys. 44, 326 (1927).

Provenance. All equations are standard textbook results. A verifier should: (1) reproduce the Robertson derivation from Cauchy–Schwarz as in Griffiths §3.5, Sakurai §1.4, Shankar §9.3; (2) confirm $[x,p]=i\hbar$ gives $\sigma_x\sigma_p\ge\hbar/2$ ; (3) check the Gaussian saturation case. For this edition all three were machine-executed (numpy/sympy), plus this article's additions: 300 random Hermitian-pair trials confirming the chain variance-product $\ge$ Schrödinger $\ge$ Robertson; reality/imaginarity of the anticommutator/commutator expectations; the spin exhibit on $|{+z}\rangle$ (spreads $\hbar/2$ each, floor $\hbar^2/4$ , exact saturation, real $\lambda = 1$ in the saturation condition, covariance zero); the Gaussian saturation ODE and its symbolic spreads ( $\sigma_x\sigma_p = \hbar/2$ for every width); the ISW ground-state product $\hbar\sqrt{\pi^2/12 - 1/2} = 0.5678\hbar$ recomputed symbolically (ownership of the ladder remains with the wave chapter's standard-deviation article and the ISW article); and the spin-1 $m{=}0$ vanishing-floor exhibit. Per the chapter ownership map: Cauchy–Schwarz is owned by Inner Product and Norm of States (consumed, not re-derived); the variance norm form by Expectation Values and Variance; $[\hat x,\hat p] = i\hbar$ is quoted here and derived at Canonical Commutation Relation; the energy–time relation is NOT a corollary of this theorem (time is a parameter, not an operator) and is owned by The Energy-Time Uncertainty Relation and Mandelstam-Tamm Bound and Evolution Timescales in the dynamics chapter.

Builds on: Hermitian Operators and Observables · Inner Product and Norm of States · Dirac Bra-Ket Notation · Born Rule and Measurement Probabilities · Linear Operators on Hilbert Space

Leads to: Completeness Relation and Resolution of Identity · Eigenvalues and Eigenstates · The Postulates of Quantum Mechanics

Where does the energy–time uncertainty relation fit into this theorem?

It doesn't — not directly, and the distinction is worth owning. Robertson's bound needs two operators; time in quantum mechanics is a parameter, not an observable (there is no self-adjoint $\hat t$ conjugate to a bounded-below $\hat H$ — Pauli's argument). The legitimate $\Delta E\,\Delta t \gtrsim \hbar/2$ statements bound evolution timescales by energy spread — how long a state needs to become distinguishable from itself — and are owned by the dynamics chapter's energy–time and Mandelstam–Tamm articles. Quoting equation (3) with $x \to t$ is the single most common misuse of this theorem.

Does the uncertainty principle forbid measuring x and p simultaneously on one particle?

The theorem is silent about it — it constrains preparation spreads, full stop. Joint measurement of incompatible observables is a real and subtle topic: exact joint sharp measurement is impossible (no joint eigenbasis — the commutator article), but approximate joint measurements exist and their accuracies obey their own tradeoff relations, POVM machinery living in the entanglement chapter. Three different "uncertainty" claims — preparation, joint measurability, disturbance — with three different theorems; this article proves the first.

Why does the imaginary part of ⟨f|g⟩ get to be called "the quantum part"?

Because its classical counterpart is identically zero. For commuting (classical) quantities the product $\Delta A\,\Delta B$ is symmetric — covariance is the whole overlap, and classical statistics has exactly the covariance-only version of Schrödinger's inequality (it is the positivity of the covariance matrix). The antisymmetric piece exists only because operators fail to commute; it is fixed by algebra rather than by the state's correlations, which is why no cleverness of preparation can remove it.

Is there a triple or N-observable version of the bound?

Yes, several. The clean route: the covariance matrix of $N$ centered observables plus $\tfrac{i}{2}$ times the commutator matrix must be positive semidefinite — the Robertson–Schrödinger relation is its $2\times2$ principal minor. For angular momentum triples, this yields relations like $\sigma_{L_x}^2 + \sigma_{L_y}^2 \ge \hbar|\langle L_z\rangle|$ (additive rather than multiplicative). Entropic uncertainty relations — bounding the sum of Shannon entropies of two observables' distributions — generalize differently and power modern quantum cryptography proofs; both directions are flagged rather than developed at this register.

If the Gaussian saturates at every width σ, what breaks when the packet evolves?

Free evolution preserves Gaussianity but correlates $x$ and $p$ : the packet develops the position–momentum covariance of the spreading article, the Schrödinger term grows, and $\sigma_x\sigma_p$ climbs above $\hbar/2$ even though the state stays "Gaussian." Robertson equality holds only at the waist. The states that stay minimum-uncertainty forever are the coherent states of the harmonic oscillator — the potential's refocusing exactly cancels the correlation growth, the story the coherent-states article told from the ladder-operator side.

Practically, when should I reach for Schrödinger's form instead of Robertson's?

Whenever your state has $x$ – $p$ (or quadrature) correlations: tilted/rotating Gaussians, squeezed states, wave packets mid-flight, thermal states of coupled modes. Diagnostic: compute the covariance $\tfrac12\langle\{\Delta\hat A,\Delta\hat B\}\rangle$ ; if it is comparable to the commutator term, Robertson is loose by a large factor and optimization arguments (e.g. "which state minimizes σ_x σ_p subject to constraints") will mislead. Quantum-optics homework and LIGO noise budgets are Schrödinger-form territory; textbook ground-state problems rarely are.

Does the bound survive for mixed states?

Yes. Replace every expectation by $\mathrm{Tr}(\rho\,\cdot)$ and the proof goes through with the same three moves (Cauchy–Schwarz holds in the Hilbert–Schmidt form). Mixing generally adds classical spread on top of quantum floors, so mixed states sit further from saturation: purity is a prerequisite for touching $\hbar/2$ . The density-operator article closing this chapter supplies the machinery; squeezing experiments confirm the purity requirement daily.