Sharpness has to be paid for

A wave with a single exact wavelength is a sine wave, and a sine wave stretches from one end of the axis to the other. It has no position at all. Anything that is somewhere has to be built out of more than one wavelength, and the narrower the thing, the more wavelengths it takes.

Nothing quantum has been used to draw either panel. The relation between the width of a packet and the width of its spectrum is a theorem about adding sine waves, it was known to communications engineers before quantum mechanics existed, and it is the reason a short radio pulse occupies a wide band.

The trade, stated exactly #

Let σ_x be the standard deviation of the packet’s intensity along the axis, and σ_k the standard deviation of the weights on wavenumber. Then

$$\sigma_x \, \sigma_k \ge \tfrac{1}{2},$$

with equality only for a Gaussian. The figure is drawn with Gaussian weights, so it sits exactly on the bound, and the printed product reads 0.500 for every spread the generator is given.

Now add one thing from physics: de Broglie’s relation, p = ħk. Multiplying the wavenumber spread by ħ turns it into a momentum spread, and the theorem about waves becomes

$$\sigma_x \, \sigma_p \ge \frac{\hbar}{2}.$$

That is the whole derivation. The inequality is mathematics; ħ is the exchange rate between the mathematics and the world.

Why the popular explanation is wrong #

The standard story is Heisenberg’s microscope: to see where an electron is, light must be bounced off it, and the light’s photon kicks it, so measuring position disturbs momentum.

The story is not nonsense — the disturbance is real and the arithmetic works out to about the right size — but it describes a different thing, and taking it as the explanation leads to a specific wrong belief: that the electron has a definite position and momentum which the clumsiness of the measurement prevents access to.

The figure above contains no measurement. There is no observer, no photon and no disturbance in it, and the trade is already exact. A packet with a narrow spread of wavenumbers is long, and it is long before anyone looks at it, in the same way that a pure tone lasts a long time before anyone listens to it. What the relation constrains is what a state can be, not what an experiment can find out.

The distinction has been made sharp experimentally. Measurement-disturbance relations of the Heisenberg-microscope kind are separate statements from the σ_xσ_p bound, they have a different mathematical form, and versions of the naive disturbance relation have been violated in the laboratory while the preparation relation held exactly. Two different facts wearing one name for eighty years.

The same relation everywhere else #

Because it is a theorem about waves rather than about particles, it turns up in every subject with a wave in it, and the counterparts are worth listing because they make it unsurprising.

Radio. A transmission lasting Δt occupies a bandwidth of at least about 1/Δt. This is why a fast data rate needs a wide channel, why an amplifier’s rise time and its bandwidth are inversely related, and why spectrum is allocated by bandwidth rather than by frequency alone.

Music. A very short note has no definite pitch. Striking a woodblock produces a click; the same energy stretched over a second produces a tone. An oscilloscope-length burst of a “440 Hz” tone occupies a band tens of hertz wide, and no filtering can narrow it without lengthening the burst.

Optics. A narrow slit spreads a beam by an angle inversely proportional to its width. That is the same relation with position across the slit and transverse momentum as the pair — and the single-slit diffraction pattern is literally a measurement of it.

The same relation is an optics experiment anybody can set up. Narrowing a slit localises the wave more precisely across the beam and widens the spread of directions it leaves in — at two wavelengths wide the pattern is a broad fan, at twenty it is a narrow one, and the product of the two widths does not fall below a bound. Position across the aperture and transverse momentum are a conjugate pair, and the pattern on the screen is the bound being paid in public.

Spectroscopy. A state that lasts τ has an energy width of about ħ/τ, which is why an atomic line has a natural width and why the sharpest transitions are the ones between long-lived states.

What it buys, rather than what it forbids #

The relation is usually presented as a limitation. Several of its most important consequences are the opposite.

Atoms have a size. Confining an electron to a small region forces a large momentum spread and therefore a large kinetic energy. Minimising the sum of that cost and the Coulomb gain gives the Bohr radius, so the relation is what stops the atom collapsing.

Solids have a volume. The same argument applied to matter under pressure gives the degeneracy pressure that resists compression, which combined with the exclusion principle is why a rock is hard.

Nothing is ever at rest. A harmonic oscillator’s ground state has energy ħω/2 rather than zero, because a state with both position and momentum exactly zero would violate the bound. That residual motion is measurable: it broadens X-ray diffraction peaks at absolute zero, it is why helium does not freeze under its own vapour pressure, and it is the reason a chemical bond’s length has a spread even in a crystal at 0 K.

Empty space is not empty. Applying the relation to a field rather than to a particle gives fluctuations in a vacuum whose consequences — the Casimir force, the Lamb shift, spontaneous emission itself — are measured to many decimal places.

Measuring the product off the drawing #

The claim that σ_xσ_k = ½ for a Gaussian is easy to assert and worth checking against the picture, because a figure that agrees with its own caption by construction proves nothing.

The generator behind the packet draws two brackets — one across the packet at ±σ_x, one across the spectrum at ±σ_k — and prints each width. The site’s figure gate then reads those brackets back through each panel’s own axis ticks, in pixels, and requires the printed widths to match the drawn ones and their product to be a half. It does this at three different spreads, and separately checks that increasing the spectral width narrows the packet by the same factor.

That arrangement is the site’s standing habit rather than a flourish, and here it earns its place for a specific reason. The relation is an inequality, and an inequality is exactly the kind of claim a figure can appear to satisfy while drawing something else — a packet that is too wide still satisfies σ_xσ_k ≥ ½, and no visual inspection would catch it. Only requiring equality, and requiring it at several spreads, tests that the drawn packet is the minimum one the caption says it is.

A discrete sum repeats itself #

The continuum of wavenumbers in the figures above is an idealisation, and swapping it for a finite set changes the picture in a way that connects this page to the standing waves the field started with.

That recurrence is not an artefact of the drawing. It is the same fact that makes a plucked string’s shape return after one period, that makes a digitally sampled signal alias, and that gives a particle in a ring a set of revivals rather than a single spreading pulse. A localised object built from a discrete spectrum is never localised only once.

The continuum limit — infinitely many wavenumbers, infinitesimally spaced — pushes the recurrences infinitely far away, and that is what “a free particle” means. Confinement of any kind discretises the spectrum and brings the recurrences back into view.

The energy–time relation is a different animal #

ΔEΔt ≥ ħ/2 is written in the same form and does not mean the same thing, and the difference is worth being careful about because it is a common error.

Position and momentum are both observables, and both σ’s are standard deviations of measurement outcomes. Time is not an observable in the same sense — there is no operator for it, and a system does not have a “time” the way it has a momentum. Δt in the energy–time relation is a characteristic timescale over which something about the system changes, and different derivations give it different precise meanings: the lifetime of a decaying state, the time for an observable to shift by one standard deviation, the duration of a pulse.

Each of those is a real and useful statement. None of them is “energy conservation may be violated for a short time”, which is a phrase in wide circulation and is not what any of the derivations say. Energy is conserved exactly; what a short-lived state lacks is a definite energy, which is a statement about the width of a distribution rather than a licence to borrow.

What it costs to know both anyway #

The bound constrains a single system prepared once. It does not forbid learning both quantities to arbitrary precision about an ensemble, and the distinction is what makes the relation testable at all.

The measurement of σ_xσ_p is made by preparing many identical systems, measuring position on some and momentum on others, and taking the two distributions’ widths. No single system is measured twice; nothing is disturbed twice. The number that comes out is a property of the preparation procedure, and it is bounded below by ħ/2 however careful the apparatus is.

That framing also explains why the relation has practical teeth in metrology. An interferometer measuring a mirror’s position more precisely necessarily imparts a larger momentum spread, which shakes the mirror and degrades the next measurement — the standard quantum limit, and the reason gravitational-wave detectors are designed around it rather than around their lasers’ power alone.

The way past it is not to beat the inequality but to choose which side to pay on. A squeezed state narrows one width below the symmetric case and widens the other to compensate, so an experiment that only cares about one quadrature can buy precision in it with noise in the other. LIGO has run with squeezed light since 2019 and gains a real improvement in sensitivity from an inequality it never violates.

The pulse that has to be broad to be short #

The relation between duration and bandwidth is the same theorem with time and frequency as the pair, and it is the constraint that the whole of ultrafast optics is arranged around.

For a Gaussian pulse the product of the duration and the bandwidth, both measured at half maximum, is at least 0.441. A pulse a hundred femtoseconds long therefore needs at least four and a half terahertz of bandwidth, which at eight hundred nanometres is about nine nanometres of wavelength — a modest requirement, met by many laser materials.

Ask for five femtoseconds and the requirement becomes ninety terahertz, which is a quarter of the carrier frequency itself: the pulse spans roughly seven hundred to a thousand nanometres, and no material with a narrow gain band can produce it. That is why titanium-doped sapphire, whose gain extends over four hundred nanometres, displaced every earlier laser medium in this field. The bandwidth is not a convenience; it is the pulse.

Push further and the inequality forces a change of carrier. A single cycle at eight hundred nanometres lasts 2.7 femtoseconds, and nothing shorter exists at that wavelength — a pulse cannot be shorter than one oscillation of the light in it. Getting into the attosecond range therefore requires light of a much shorter wavelength, which is why attosecond pulses are made in the extreme ultraviolet, by driving a gas so hard that it radiates high harmonics of the driving field. The drive toward shorter pulses is a drive toward shorter wavelengths, and the inequality is what makes it so.

The same theorem run backwards is the reason optical frequencies can be measured at all. A train of pulses repeating steadily has a spectrum of sharp teeth spaced by the repetition rate, and the width of each tooth is set by how long the train lasts. Lock such a train for a second and each tooth is about a hertz wide out of four hundred terahertz — a part in $10^{14}$, which is a ruler fine enough to count optical cycles against a caesium clock. The frequency comb is that arrangement, it made optical clocks possible, and it is the long-pulse end of exactly the inequality whose short-pulse end forces a laser to be broadband.

One further technique is worth naming because it exploits the essay’s own remark about phase. Amplifying a femtosecond pulse directly destroys the amplifier, because the peak power is enormous. Chirped-pulse amplification stretches the pulse in time first — by sending it through a dispersive element that delays the components differently, leaving the spectrum untouched and only rearranging the phases — amplifies the now-harmless long pulse, and then undoes the delay to recompress it. Nothing about the bandwidth changes at any stage; what is manipulated is the phase relation the essay says decides whether a packet is at its minimum width. That trick is the basis of every high-power short-pulse laser built since the mid-1980s.

The correction Bohr made and nobody kept #

The microscope argument is not merely a later simplification. It is in Heisenberg’s original paper, and it was criticised as wrong within weeks by the person best placed to criticise it.

Heisenberg published the relation in 1927 with the microscope as its physical justification: to locate an electron, light must be scattered from it, the shorter the wavelength the better the location, and the shorter the wavelength the harder the recoil. Bohr, returning to Copenhagen shortly afterwards, objected on two grounds.

The first was technical and simply an error: Heisenberg’s treatment of the microscope’s resolving power was wrong, and correcting it changes the argument’s arithmetic. The second was the substantive one. Bohr held that the relation does not arise from a disturbance at all, but from the fact that a wave description and a particle description are both needed and cannot be used at once — so the limitation is in what states exist rather than in what a measurement does to them.

The disagreement was severe, and it delayed publication. Heisenberg added a note in proof acknowledging Bohr’s criticism and stating that the relations can be obtained without appealing to any discontinuous change caused by the observation.

That note is in the paper, and the microscope is what everybody remembers. Eighty years of textbooks have carried the disturbance reading, which the author’s closest collaborator had rejected before the ink was dry.

The modern version of the correction is quantitative. Ozawa showed in 2003 that the naive error-times-disturbance inequality is not universally valid and derived a corrected relation with additional terms. Two experiments in 2012 — one on neutron spins, one on photon polarisation — measured error and disturbance separately and found the naive product below the supposed bound while the corrected relation held.

The situation is not entirely settled, and the reason is worth stating honestly: it turns on how “error” and “disturbance” are defined, and other formulations of those quantities give inequalities that are not violated. What is not in dispute is the essay’s central point. The preparation relation between the widths of position and momentum distributions is exact, holds with no measurement anywhere, and is a different statement from anything about what an apparatus does to a system.

What the figure cannot show #

The packet drawn is real-valued and symmetric, and both are special.

A general wavefunction is complex, and the phase relationship between its components is what decides whether the packet is at its minimum width or spread far beyond it. A packet that has been allowed to propagate acquires a phase curvature — different wavenumbers travel at different speeds if the medium is dispersive, which for a matter wave it always is — and it broadens without any of its spectral content changing. So the figure shows a packet at one instant and hides the fact that a free particle’s position spread grows with time while its momentum spread does not.

The other half of the story belongs to an earlier rung: once the components travel at different speeds, the packet moves at one speed and the ripples inside it at another, and the envelope spreads as it goes. Every quantum packet does this, and it means the drawing above is a snapshot taken at the one moment when the spreading is least. A moment later the position width is larger and the momentum width is unchanged, so the product has risen — the bound is saturated once and then only exceeded.

The figure also shows a one-dimensional packet, and the relation constrains each pair of conjugate directions separately — position along x against momentum along x, and nothing at all about the momentum along y. Two quantities can both be sharp if they are not conjugate, and knowing which pairs are conjugate is a structural fact about the theory rather than a general suspicion that everything is fuzzy.

Why nothing has a trajectory #

The relation is the sharpest available statement of what has been given up, and it is worth naming plainly because the rest of the field depends on it.

A trajectory is a position at every instant, and a position at every instant implies a velocity at every instant. The bound says a state with both sharply defined does not exist — not that it exists and cannot be found, but that the description is not available. So the classical picture the site is built out of, in which a body has a place and a speed and the two evolve, is not a simplification of the quantum one. It is a different kind of object.

What replaces it is a state that evolves smoothly and deterministically, and whose relationship to any particular measurement is a distribution. Determinism survives; the trajectory does not. That is a genuinely odd combination and it is why the field’s remaining rungs are largely about the second half: what a measurement does, why one outcome rather than another, and how the classical picture is recovered for anything large.

There is a numerical way to feel the boundary. A dust grain of a microgram, localised to a micrometre, has a momentum spread corresponding to a velocity spread of 5 × 10⁻²² metres per second — it would take longer than the age of the universe for that to smear its position by an atomic diameter. An electron localised to an atom has a velocity spread of a million metres per second. The relation is the same; the consequences are separated by forty orders of magnitude.

Where the ladder goes next #

The rungs from here: the commutator, and the general relation that gives a bound for any pair of observables from the algebra rather than from Fourier analysis; squeezed states, in which one width is deliberately pushed below the symmetric case at the other’s expense — used in gravitational-wave detectors to beat the shot-noise limit; the energy–time relation done carefully, and the several different theorems it names; the exclusion principle and degeneracy pressure, where the bound becomes a mechanical force; and the measurement problem itself, where the question of what a measurement does is separated from the question of what a state can be.

The claim to carry forward is that this is not a statement about ignorance and not a statement about clumsiness. A narrow packet is made of many wavelengths, and that is arithmetic. The physics enters in one place only, when ħ converts a wavenumber into a momentum, and everything strange about the result was already true of sound.