The Energy-Time Uncertainty Relation

Every physics student can recite it: $\Delta E\,\Delta t \gtrsim \hbar/2$ . It is quoted to explain why spectral lines have width, why short laser pulses are broadband, why particle physicists call a resonance's width a lifetime in disguise. And yet, of all the famous inequalities in quantum mechanics, this is the one most often stated wrongly — because it looks like the position-momentum relation while resting on entirely different ground. In this theory $\hat x$ and $\hat p$ are operators; time is not. There is no $\hat t$ , no $[\hat t, \hat H] = i\hbar$ , no "spread of measured times" — and Pauli proved there cannot be, so long as energies have a floor. What survives is better than the folklore: a ladder of precise readings, each a theorem in its own domain. At the bottom, pure wave arithmetic — a signal lasting a time $\Delta t$ must span a frequency band $\Delta\omega \gtrsim 1/\Delta t$ , and $E = \hbar\omega$ dresses that in energy units. One rung up, the lifetime-linewidth relation: a state that decays in time $\tau$ radiates a line of width $\Gamma = \hbar/\tau$ . At the top, the rigorous version — Mandelstam and Tamm's 1945 theorem, which defines $\Delta t$ as the time any chosen observable needs to change by one standard deviation and derives the bound from the uncertainty principle this chapter already owns. This article climbs that ladder and polices its rungs: what each statement says, what it does not, and where the folklore goes wrong. The summit — the Mandelstam-Tamm derivation and the quantum speed limit — is stated here and proven in the next article.

Key takeaways

$\Delta E\,\Delta t \gtrsim \hbar/2$ links a state's energy spread to how fast anything observable about it can change — but $\Delta t$ is not an operator uncertainty, because time is a parameter in quantum mechanics.
Pauli's argument: a self-adjoint $\hat t$ conjugate to $\hat H$ would force the energy spectrum to run over the whole real line — incompatible with a ground state. No time operator exists.
Reading 1 (wave arithmetic): a finite wave train of duration $\Delta t$ carries a frequency band $\Delta\omega \gtrsim 1/\Delta t$ ; Gaussians saturate with $\sigma_E \sigma_t = \hbar/2$ exactly.
Reading 2 (lifetime-linewidth): an exponentially decaying state radiates a Lorentzian line of FWHM $\Gamma = \hbar/\tau$ — verified daily in every spectroscopy lab ( $\tau = 16.2$ ns for Na 3p gives $\Delta\nu = 9.8$ MHz).
Reading 3 (the rigorous one): Mandelstam-Tamm define $\Delta t_A = \Delta A / |d\langle A\rangle/dt|$ — the time for $\langle A \rangle$ to move by one standard deviation — and prove $\Delta E\,\Delta t_A \ge \hbar/2$ . Stated here; derived at Mandelstam-Tamm Bound and Evolution Timescales.
A stationary state has $\Delta E = 0$ and infinite $\Delta t$ : nothing observable moves, ever — the relation's trivial and exact boundary case.
Key idea: time is a parameter, not an operator — every honest energy-time relation bounds an evolution timescale of the system by its energy spread, never the precision of a clock.

The core insight

Time is a parameter, not an operator. The relation $\Delta E\,\Delta t\ge\hbar/2$ is not a commutator bound but the statement that $\Delta t=\Delta A/|\mathrm{d}\langle A\rangle/\mathrm{d}t|$ — the time for any observable to change by one standard deviation — is bounded below by $\hbar/(2\Delta E)$ .

Where this lives today

The relation is production machinery at both ends of the timescale. At the short end, attosecond science: a 100-as pulse requires tens of electronvolts of coherent bandwidth (a 10 fs, 800 nm pulse already needs $\approx 94$ nm of spectrum, machine-checked), which is why attosecond sources are high-harmonic XUV combs. At the long end, optical atomic clocks invert the logic: strontium's clock transition lives over a hundred seconds, hence a millihertz natural linewidth and $Q$ above $10^{17}$ — the world's best timekeeping is the engineering of states with the smallest honest $\Delta E$ . And every "width" column in the particle data tables is a lifetime measured through equation (3), exactly as in Example 2.

The physical picture

Think of an energy eigenstate as a clock whose hands never visibly move. Its only time dependence is the global phase $e^{-iEt/\hbar}$ — a rotation nobody can read (Stationary States and Energy Eigenstates owns that theorem). To make the state visibly tick, superpose at least two energies: the relative phase turns at the Bohr frequency $\omega = \Delta E/\hbar$ , and expectation values beat at exactly that rate — the first observable change arrives a time of order $\hbar/\Delta E$ later. Energy spread is not a nuisance here; it is the fuel of change. No spread, no dynamics.

Figure

Only energy spread makes a state tick — one eigenstate is frozen, a superposition beats

The second picture is inheritance. Classical signal theory has always known the bandwidth theorem: a pulse of duration $\Delta t$ cannot be monochromatic — its spectral width obeys $\Delta\omega\,\Delta t \gtrsim 1$ . Quantum mechanics adds one dictionary entry, $E = \hbar\omega$ , and the classical theorem becomes an energy-time relation. No operators, commutators, or measurement enter at this rung — which is exactly why it is safe: pure wave arithmetic, immune to over-interpretation about clocks. The danger begins when the relation is read as a twin of $\sigma_x \sigma_p \ge \hbar/2$ . That inequality — the Robertson bound of Generalized Uncertainty Principle — needs two operators and a commutator. Time has no operator. The honest quantum content of $\Delta E\,\Delta t$ lives one level deeper: in how fast the state moves through Hilbert space, which is what the Mandelstam-Tamm construction finally pins down.

Figure

The meaning ladder — three theorems and one impostor

Why there is no time operator

The cleanest way to see what $\Delta t$ cannot mean is to try to build the operator and watch it break. Suppose a self-adjoint $\hat t$ existed with the canonical pairing to energy. The pairing would read

[\hat t, \hat H] = i\hbar \qquad (1)

where: this is the exact structural twin of $[\hat x, \hat p] = i\hbar$ — and that twin-ship is the trap. For position and momentum, the commutator is consistent because both spectra run over the whole real line. For energy, it is fatal: real Hamiltonians have a ground state, and the derivation below shows equation (1) is incompatible with any energy floor. Pauli parked this observation in a footnote of his 1933 Handbuch article, and it has policed the subject ever since.

Derivation APauli's argument — a time operator would abolish the ground state

We show that a self-adjoint operator $\hat t$ obeying equation (1) forces the spectrum of $\hat H$ to be the entire real line, unbounded below — so no system with a ground state can host one. The strategy is to use $\hat t$ as a generator of energy translations, exactly the way Canonical Commutation Relation used $\hat p$ (via the Weyl form) to generate position translations: conjugating $\hat H$ by $e^{-i\varepsilon\hat t/\hbar}$ shifts every energy by $\varepsilon$ , for any real $\varepsilon$ . Watch how the argument uses nothing but the commutator and unitarity — its generality is the point.

Step 1. Build the candidate unitary. For real $\varepsilon$ , the operator $\hat V_\varepsilon = e^{-i\varepsilon \hat t/\hbar}$ is unitary because $\hat t$ is self-adjoint (the generator doctrine of Unitary Operators and Transformations).

Step 2. Conjugate the Hamiltonian. The commutator (1) telescopes through the exponential series — the same one-rung ladder $[\hat H, f(\hat t)] = -i\hbar f'(\hat t)$ used at Canonical Commutation Relation — to give

\hat V_\varepsilon^\dagger\, \hat H\, \hat V_\varepsilon = \hat H + \varepsilon

Why this is the kill shot: conjugation by a unitary never changes a spectrum (the invariance card of Unitary Operators and Transformations). But the right-hand side has every eigenvalue shifted by $\varepsilon$ . Both statements can hold only if the spectrum is invariant under translation by every real $\varepsilon$ .

Step 3. Read the contradiction. A translation-invariant spectrum is the whole real line: for every eigenvalue $E$ there is another at $E - 10$ eV, another at $E - 10^{6}$ eV, without end. No ground state, no stable atoms, no thermodynamics. Conclusion: on a Hilbert space whose Hamiltonian is bounded below, no self-adjoint $\hat t$ satisfying (1) exists. Whatever $\Delta t$ means in the energy-time relation, it is not the standard deviation of a time observable. Verified: the finite-dimensional cousin is checked numerically — $\mathrm{tr}[\hat T, \hat H] = 0$ for any two Hermitian matrices, so $[\hat T,\hat H] = i\hbar\hat 1$ is impossible in any finite dimension ( $10^{-12}$ ); and the shift mechanism itself is exhibited on a Fourier grid, where multiplying by $e^{iax}$ translates the momentum spectrum by exactly $a$ (verify script).

The escape routes are worth naming, because research uses them: give up self-adjointness and $\hat t$ survives as a positive-operator-valued measure (the arrival-time observables of quantum optics); give up the floor and equation (1) is consistent. Inside standard quantum mechanics the verdict stands: time is a label in the Schrödinger equation, and the relation we are studying must be a statement about dynamics, not about clocks.

Rung 1 — the bandwidth theorem in energy units

Strip away all quantum machinery and consider any amplitude $f(t)$ that lives for a finite while — a pluck, a pulse, a wave train. Its frequency content is its Fourier transform, and the widths of the two profiles are chained to each other. The quantum version needs one substitution: a state's energy distribution is carried by amplitudes $e^{-iEt/\hbar}$ , so frequency width becomes energy width through $E = \hbar\omega$ . For Gaussian profiles the chain is an exact equality,

\sigma_E\, \sigma_t = \frac{\hbar}{2} \qquad (2)

where: $\sigma_t$ is the standard deviation of $|f(t)|^2$ (the duration of the amplitude) and $\sigma_E = \hbar\sigma_\omega$ the standard deviation of its spectral weight. The factor $\hbar/2$ with equality marks the Gaussian as the transform-limited waveform — the same extremal role Gaussians play for $\sigma_x\sigma_p$ in Generalized Uncertainty Principle, and for the same mathematical reason. Every other shape does worse. This is the rung on which femtosecond optics lives, and the momentum-space dictionary of Momentum-Space Wavefunction is its spatial sibling.

Go deeper · Derivation B · 1 minThe bandwidth floor and its Gaussian saturation

Derivation BThe bandwidth floor and its Gaussian saturation

We derive the time-frequency trade-off for a Gaussian wave train and exhibit the finite-train (sinc-squared) case, which is the shape every finite observation actually produces. The goal is to see that the floor is pure Fourier analysis — no operators, no measurement postulate — and to put honest numbers on the "sharp turn-on" case that textbooks wave at. The Gaussian computation is exact; the truncated train yields the universal $\mathrm{FWHM} \times T \approx 5.57$ product that spectroscopists internalize.

Step 1. Take the Gaussian amplitude $f(t) = e^{-t^2/4\sigma^2}$ , whose intensity $|f|^2 = e^{-t^2/2\sigma^2}$ has standard deviation exactly $\sigma_t = \sigma$ (moment integral, machine-checked symbolically).

Step 2. Fourier transform. The transform of a Gaussian is a Gaussian, $\tilde f(\omega) \propto e^{-\sigma^2\omega^2}$ , whose spectral intensity has standard deviation $\sigma_\omega = 1/2\sigma$ . The product is width-independent:

\sigma_\omega\,\sigma_t = \frac{1}{2} \quad\Longrightarrow\quad \sigma_E\,\sigma_t = \frac{\hbar}{2}

Why equality here and inequality in general: the Cauchy-Schwarz step that bounds all other shapes is saturated only by Gaussians — the same extremal logic as the minimum-uncertainty wave packets of Generalized Uncertainty Principle.

Step 3. The finite train. Turn a pure oscillation $e^{-i\omega_0 t}$ on for a window $T$ and off again. Its spectrum is

|\tilde f(\omega)|^2 = T^2\,\mathrm{sinc}^2\!\Big(\frac{(\omega - \omega_0)T}{2}\Big), \qquad \mathrm{FWHM}(\Delta\omega) = \frac{5.566}{T}

so a sharp-edged observation of duration $T$ smears energies by $\Delta E_{\rm FWHM} \approx 5.57\,\hbar/T$ — worse than the Gaussian floor, never better. Every finite-time measurement inherits this resolution limit. Verified: Gaussian moments and the product $\sigma_\omega\sigma_t = 1/2$ exactly (sympy); the sinc half-power point $x = 1.39156$ and $\mathrm{FWHM}\times T = 5.566$ by root-finding; the sinc-squared form against direct numerical Fourier integrals at three detunings ( $10^{-10}$ ) — verify script.

Rung 2 — lifetime and linewidth

An excited atom does not oscillate forever; its amplitude decays. Model the emitted amplitude as an oscillation that dies exponentially, $f(t) = \theta(t)\,e^{-i E_0 t/\hbar}\,e^{-t/2\tau}$ , so that the probability of still being excited falls as $e^{-t/\tau}$ . What line does this radiate? Two lines of Fourier arithmetic answer, and the answer is the most measured consequence of the energy-time relation:

I(E) \propto \frac{1}{(E - E_0)^2 + (\Gamma/2)^2}, \qquad \Gamma = \frac{\hbar}{\tau} \qquad (3)

where: the profile is a Lorentzian centered on the transition energy $E_0$ , with full width at half maximum $\Gamma$ fixed by the lifetime alone. This is the natural linewidth — the irreducible fuzziness of any spectral line, present with the atom at rest, in vacuum, at zero temperature. The calculation was a Fourier transform of a decaying signal — half a line of algebra, verified symbolically (the half-maximum lands at exactly $E_0 \pm \Gamma/2$ ). The deeper questions — why decay is exponential, where that law comes from spectrally, when it fails — belong to Survival Amplitude and Decay of Unstable States; and the subtlety that a Lorentzian has no finite energy variance (so Mandelstam-Tamm says nothing directly about it) is confronted at Mandelstam-Tamm Bound and Evolution Timescales. The optical-potential model of Conservation of Normalization is this rung's operator-level cousin: an anti-Hermitian $-i\Gamma/2$ in the Hamiltonian drains the norm on the same clock $\tau = \hbar/\Gamma$ — its nuclear-resonance number ( $\Gamma = 6$ MeV $\Rightarrow \tau = 1.1\times10^{-22}$ s) is equation (3) read backwards.

Rung 3 — the rigorous statement, and its owner

The top rung repairs the operator-shaped hole. Since no $\hat t$ exists, Mandelstam and Tamm (1945) changed the question: fix any observable $\hat A$ and ask how long its mean needs to shift by one standard deviation — a statistically resolvable change. That time is

\Delta t_A \;\equiv\; \frac{\Delta A}{\big|\,\mathrm{d}\langle \hat A\rangle/\mathrm{d}t\,\big|} \qquad (4)

where: $\Delta A$ is the ordinary quantum spread of Standard Deviation and Quantum Uncertainty, and the denominator is the drift rate of the mean — the object the expectation-value engine of this chapter's Ehrenfest's Theorem computes as $(1/i\hbar)\langle[\hat A,\hat H]\rangle$ . The theorem then reads: for every such observable, in every state,

\Delta E\;\Delta t_A \;\ge\; \frac{\hbar}{2} \qquad (5)

where: $\Delta E$ is the state's energy spread. Small energy spread means every observable drifts slowly compared to its own noise; and a specialization to the survival probability yields the quantum speed limit $\tau_\perp \ge \pi\hbar/(2\Delta E)$ — a state with $\Delta E = 1$ eV needs at least $1.03$ fs to evolve into anything orthogonal. We state both results and stop: the derivations — the Robertson-plus-Ehrenfest chain, the arccosine survival bound, the two-level state that saturates them, and the companion Margolus-Levitin bound — are the entire business of the next article, Mandelstam-Tamm Bound and Evolution Timescales. (That the Robertson inequality alone cannot produce (5) by "substituting $t$ for $x$ " is precisely the misuse Generalized Uncertainty Principle flags and defers here.) The boundary case makes the meaning vivid. In an energy eigenstate $\Delta E = 0$ — and indeed nothing moves: every $\langle\hat A\rangle$ is frozen (checked to $10^{-13}$ for random observables), so $\Delta t_A = \infty$ . Stationary states do not threaten the relation; they anchor it.

Limiting cases

case	result	meaning
stationary state	$\Delta E = 0 \Rightarrow \Delta t \to \infty$	eigenstates never tick; phase alone turns (Stationary States and Energy Eigenstates)
two-level beat	$\Delta t \sim \hbar/\Delta E$	Bohr-frequency beats — the minimal ticking clock (Spectral Decomposition of Time Evolution)
decaying state	$\Gamma\,\tau = \hbar$ , Lorentzian line	natural linewidth, equation (3); full decay theory at Survival Amplitude and Decay of Unstable States
transform-limited pulse	$\sigma_E\sigma_t = \hbar/2$	Gaussian saturates the bandwidth floor — equation (2)
sharp-edged observation, duration $T$	$\Delta E_{\rm FWHM} \approx 5.57\,\hbar/T$	finite windows smear energy — the sinc-squared tax
conserved observable	$[\hat A,\hat H] = 0 \Rightarrow \mathrm{d}\langle A\rangle/\mathrm{d}t = 0$	$\Delta t_A = \infty$ for that observable — the trivial direction of Constants of the Motion and Conservation Laws

Common trapDelta t is the accuracy of your clock, or the duration of your measurement

Neither, and the difference costs marks and understanding. $\Delta t$ in every legitimate version is a property of the system and the observable you watch — the time for $\langle\hat A\rangle$ to drift by one $\Delta A$ (Mandelstam-Tamm), a wave-train duration (bandwidth theorem), or a lifetime (linewidth relation). It is not stopwatch resolution, and the relation does not say "measuring energy for a short time disturbs it by $\hbar/\Delta t$ " — the reading Einstein's photon box exploited and Bohr's rebuttal untangled (history band). Two tells you are misreading it: (i) you write $[\hat t,\hat H] = i\hbar$ — forbidden by Pauli's argument; (ii) you apply it to a single sharp eigenstate and expect motion. When in doubt, name the observable: whose $\Delta t$ is it?

Worked examples

Example 1 — the sodium D-line's natural width (confidence builder). The Na $3p$ level lives $\tau = 16.2$ ns before decaying to $3s$ at 589 nm. Find the natural linewidth in energy and frequency, and the line's intrinsic quality factor.

Step 1. Energy width from the lifetime, equation (3): $\Gamma = \hbar/\tau = \dfrac{6.582\times10^{-16}\ \mathrm{eV\,s}}{16.2\times10^{-9}\ \mathrm{s}} = \boxed{4.06\times10^{-8}\ \mathrm{eV}}$ — forty nano-electronvolts riding on a 2.1 eV transition.

Step 2. Frequency width: $\Delta\nu = \Gamma/h = 1/(2\pi\tau) = \boxed{9.8\ \mathrm{MHz}}$ . This is the FWHM a perfect spectrometer would record from a single atom at rest.

Step 3. Quality factor: $\nu_0 = c/589\,\mathrm{nm} = 5.09\times10^{14}$ Hz, so $Q = \nu_0/\Delta\nu = \boxed{5.2\times10^{7}}$ . For scale, room-temperature Doppler broadening of this line is $\sim$ 1.5 GHz — 150 times wider. The natural width is the floor beneath thermal motion; reaching it is what laser cooling is for.

Example 2 — reading a particle's lifetime off a plot (exam level). The Z boson's resonance curve, measured at LEP, has width $\Gamma_Z = 2.4952$ GeV. No clock can time its decay directly. What lifetime does the lineshape imply, and why is the inference legitimate?

Step 1. Invert equation (3): $\tau_Z = \hbar/\Gamma_Z = \dfrac{6.582\times10^{-16}\ \mathrm{eV\,s}}{2.4952\times10^{9}\ \mathrm{eV}} = \boxed{2.64\times10^{-25}\ \mathrm{s}}$ .

Step 2. Legitimacy check. The Breit-Wigner resonance profile is equation (3)'s Lorentzian, arising the same way — Fourier analysis of an exponentially decaying amplitude (the resonance machinery this corpus builds at Resonant Transmission and Tunnelling Resonances and completes in the scattering chapter). Width in the energy domain and decay in the time domain are one dataset in two Fourier representations, not two measurements.

Step 3. Perspective: light crosses a proton radius in $3\times10^{-24}$ s — ten times longer than $\tau_Z$ . The Z never leaves a measurable track; its entire observable existence is the 2.5 GeV of energy blur in the lineshape. Here the energy-time relation is not a limitation — it is the measurement method.

The man who interrogated Delta t

Origins · 1 minRudolf Peierls and how this was discovered

Rudolf Peierls

1907–1995 · Zurich, then Birmingham and Oxford

Peierls was twenty-three, working with the twenty-two-year-old Lev Landau in Zurich, when the two asked the question this article polices: what does $\Delta E\,\Delta t$ actually license you to say? Their 1931 paper pushed the uncertainty principle into energy-time and relativistic territory and found the folklore reading — "energy measured over a time $t$ is uncertain by $\hbar/t$ " — far subtler than advertised: some versions survive scrutiny, others they argued were artifacts of bad operational definitions. It was the first paper to treat the energy-time relation as something to be derived and delimited rather than asserted — the program Mandelstam and Tamm completed in 1945. Peierls's later career made him one of physics' great clarifiers: the quantum theory of solids (bands, holes, the Peierls instability), the 1940 Frisch-Peierls memorandum (one line of history's heaviest arithmetic), and decades at Birmingham training a generation. His Surprises in Theoretical Physics — a book of "the obvious answer is wrong, here is the careful one" — is this article's spirit in paperback. Landau's portrait bands live elsewhere in this corpus; here he is the co-author who shared the question.

History

1927

— Heisenberg's uncertainty paper sketches an energy-time relation by analogy; Bohr presses on what Δt means

1930

— Einstein's photon-box challenge at the Solvay congress; Bohr's overnight rebuttal uses gravitational red-shift to restore ΔE Δt ≳ ℏ

1931

— Landau and Peierls publish the first critical analysis extending the uncertainty principle to energy, time, and relativistic measurement

1933

— Pauli's Handbuch footnote: no self-adjoint time operator can be conjugate to a Hamiltonian with a ground state

1945

— Mandelstam and Tamm derive the rigorous form: ΔE bounds the evolution timescale of every observable

1998

— Margolus and Levitin add the mean-energy speed limit, completing the modern quantum-speed-limit pair

The photon box deserves its paragraph, because it is the finest stress test the relation ever faced. Einstein imagined a box of radiation with a clockwork shutter open for a time $\Delta t$ , releasing one photon; weigh the box before and after, and $E = mc^2$ delivers the photon's energy exactly — beating the bound with a spring scale. Bohr's overnight reply: weighing means letting the box settle in a gravitational field, the settling displaces the clock, and the red-shift makes the clock rate itself uncertain by exactly enough to restore $\Delta E\,\Delta t \gtrsim \hbar$ . Einstein's own relativity closed his loophole — and fixed the understanding that the relation constrains dynamics and inference, not stopwatch quality.

Going further

References. Griffiths, Introduction to Quantum Mechanics, 3rd ed., §3.5.3 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §1.4 and §2.1 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 4 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Complement D_III · Mandelstam & Tamm, J. Phys. (USSR) 9, 249 (1945) · Margolus & Levitin, Physica D 120, 188 (1998).

Provenance. Definitions and inequalities cross-checked against Sakurai & Napolitano §1.4/§2.1, Griffiths §3.5.3, and Cohen-Tannoudji Complement DIII. A verifier should: (1) re-derive MT from Robertson + Ehrenfest as shown; (2) confirm $\tau_\perp=\pi\hbar/(2\Delta E)$ against Mandelstam-Tamm (1945) and the Margolus-Levitin (1998) bound; (3) check the numerical estimate using $\hbar=6.582\times10^{-16}\,\mathrm{eV\,s}$ , giving $\tau_\perp\approx1.03\times10^{-15}$ s for $\Delta E=1$ eV; (4) verify the Lorentzian-Fourier pair for exponential decay yielding $\Gamma\tau=\hbar$ . For this edition all quantitative claims were machine-verified (sympy/numpy/scipy, 24 checks green): Gaussian $\sigma_\omega\sigma_t = 1/2$ exact; truncated-train $\mathrm{FWHM}\times T = 5.566$ ; the Lorentzian half-maximum at $E_0 \pm \Gamma/2$ ; the Na 3p, Z-boson, and 10-fs-pulse numbers; the finite-dimensional trace no-go for $[\hat T,\hat H] = i\hbar$ ; the eigenstate-freezes check; and the $\sigma_H = 1$ eV $\Rightarrow \tau_\perp = 1.034$ fs anchor. Per the chapter ownership map: the Mandelstam-Tamm derivation, the arccosine survival bound, saturation, and Margolus-Levitin comparison belong to Mandelstam-Tamm Bound and Evolution Timescales (next serial — stated here only); the Robertson inequality to Generalized Uncertainty Principle (whose explicit "energy-time is not a corollary" defer this article cashes); the expectation-value engine to this chapter's Ehrenfest's Theorem; survival-amplitude theory, exponential decay's spectral origin and its failures to Survival Amplitude and Decay of Unstable States; the $\tau = \hbar/\Gamma$ optical-potential instance to Conservation of Normalization (its number quoted, not re-derived); Fourier-pair machinery to Momentum-Space Wavefunction.

Builds on: Time-Dependent Schrodinger Equation (QM3) · Ehrenfest's Theorem (QM3) · Stationary States and Energy Eigenstates · Spectral Decomposition of Time Evolution

Leads to: Mandelstam-Tamm Bound and Evolution Timescales · Constants of the Motion and Conservation Laws · The Time-Evolution Operator · The Heisenberg Equation of Motion

If time is not an operator, why does ΔE·Δt look exactly like Δx·Δp?

Because both descend from the same Fourier mathematics — but through different physics. Position-momentum is a commutator theorem: two operators, Robertson's bound, done. Energy-time is Fourier duality between a state's energy content and its evolution history, upgraded by Mandelstam-Tamm into a theorem about observables' drift rates. The resemblance is real (both trade sharpness in conjugate domains) and the difference is real (no $\hat t$ exists — Pauli). Treating the resemblance as identity is the single most common misuse of the uncertainty principle.

Does the relation mean energy conservation can be violated for short times?

No — that gloss is folklore. In unitary quantum mechanics $\langle\hat H\rangle$ is exactly conserved for a closed system (Constants of the Motion and Conservation Laws). What the relation licenses is different: a state prepared or probed over a finite window does not have a sharp energy to conserve — its distribution has width $\gtrsim \hbar/\Delta t$ . Virtual-particle talk in field theory is bookkeeping for exactly this spectral width, not a suspension of conservation laws.

Which Δt do I use on an exam problem?

Name the physics first. Decay or resonance width: $\Gamma = \hbar/\tau$ (rung 2). Pulse or wave-train duration: bandwidth theorem, $\sigma_E\sigma_t \ge \hbar/2$ with Gaussian equality (rung 1). "How fast can this state change / become distinguishable": Mandelstam-Tamm, $\Delta t_A = \Delta A/|d\langle A\rangle/dt|$ or $\tau_\perp = \pi\hbar/2\Delta E$ (rung 3, next article). If the problem hands you a lifetime, it wants rung 2; if it hands you an energy spread and asks about evolution, rung 3.

Why is the natural line shape a Lorentzian and not a Gaussian?

Because the emitted amplitude decays exponentially, and the Fourier transform of $\theta(t)e^{-t/2\tau}$ is $1/(i\Delta\omega + 1/2\tau)$ , whose modulus squared is a Lorentzian. Gaussians come from Gaussian envelopes (Doppler broadening gives them, since thermal velocities are Gaussian). Measured lines are usually Voigt profiles — a Lorentzian core convolved with a Gaussian — and disentangling the two widths is routine spectroscopic practice. Why decay is exponential in the first place, and when it is not, is the business of Survival Amplitude and Decay of Unstable States.

Can I beat the linewidth limit by measuring longer?

You can beat the instrumental limit but never the natural one — it belongs to the emitted field, not your spectrometer. Many lifetimes of watching gives the full Lorentzian, width $\Gamma = \hbar/\tau$ ; a short window $T \ll \tau$ adds sinc-squared broadening $\sim 5.57\hbar/T$ . The route to a narrower line is a longer-lived state — the strontium clock-transition strategy behind modern frequency standards.

What exactly did Einstein's photon box get wrong?

Nothing in its quantum mechanics — the flaw was forgetting general relativity. Weighing the box means letting it settle in a gravitational field; the settling displaces the clock, and the red-shift makes the clock rate uncertain in proportion to the weighing precision. Multiply out and $\Delta E\,\Delta t \gtrsim \hbar$ returns exactly — a beautiful consistency check, and a warning that "obvious" loopholes usually smuggle in an ignored piece of physics.