Global and Relative Phase

A normalized state vector is not quite a physical state. The whole family $\{e^{i\theta}|\psi\rangle:\theta\in[0,2\pi)\}$ — a ket and all its phase rotations — describes a single physical situation, indistinguishable by any experiment. This redundancy is forced by the Born rule: every prediction has the form $|\langle\phi|\psi\rangle|^2$ or $\langle\psi|A|\psi\rangle$ , and in each an overall factor $e^{i\theta}$ meets its conjugate $e^{-i\theta}$ and cancels. That untouchable factor is the global phase, and its unobservability is why pure states live in the projective Hilbert space of rays, $\mathbb{CP}^{n-1}$ — for one qubit, the Bloch sphere. A relative phase is a completely different animal. Inside a superposition $\alpha|0\rangle+\beta|1\rangle$ the amplitudes carry their own phases, and the difference $\varphi=\arg\beta-\arg\alpha$ survives every cancellation. It is the azimuth on the Bloch sphere, the content of $\rho$ 's off-diagonal coherences, and the single quantity that turns a phase into a fringe. Change $\varphi$ and you move to a genuinely different physical state; change the global phase and you have not moved at all. The deepest lesson is that the boundary between the two is contextual. A phase that is merely global for an isolated subsystem is promoted to an observable relative phase the instant that subsystem becomes one branch of a larger superposition — the mechanism behind controlled gates, phase kickback, the Aharonov–Bohm effect, and Berry's geometric phase. Getting it right is the prerequisite for every interference-based algorithm.

Key takeaways

A global phase is an overall factor $e^{i\theta}$ on the entire state vector; it cancels in every Born-rule expression and is therefore completely unobservable.
A relative phase is the phase difference between superposed amplitudes; it is fully physical because it controls interference, and it is the azimuthal angle $\varphi$ on the Bloch sphere.
Because the global phase is redundant, pure states are rays — points of $\mathbb{CP}^{n-1}$ — and the density operator $\rho=|\psi\rangle\langle\psi|$ carries only physical information.
In the computational basis both phases are invisible ( $P(0)=|\alpha|^2$ ); in a rotated basis the relative phase appears as a fringe $P(\pm)=\tfrac12(1\pm2|\alpha||\beta|\cos\varphi)$ .
A phase that is global for an isolated qubit becomes a relative phase for the whole when that qubit is a control branch — this contextual promotion is phase kickback, and it powers phase estimation and Shor's algorithm.
Key idea: All measurement statistics depend only on $|\langle\phi|\psi\rangle|^2$ and $\langle\psi|A|\psi\rangle$ ; an overall factor $e^{i\theta}$ cancels in both, but a phase difference between superposed amplitudes does not — it is the engine of quantum interference.

Where this lives today

The global/relative distinction is load-bearing engineering. Every quantum compiler works in the projective unitary group: transpilers for superconducting and trapped-ion hardware freely drop overall phases from single-qubit unitaries, yet must scrupulously keep them whenever a gate becomes a controlled operation, because kickback would otherwise be miscomputed. Quantum phase estimation — the engine inside Shor's algorithm and quantum-chemistry eigensolvers — is industrial-scale phase kickback: it converts a unitary's eigenvalue from an unobservable global phase into a string of relative phases across a register, read out by an inverse quantum Fourier transform. Ramsey interferometry, the heart of every atomic clock, prepares $|+\rangle$ , lets $\varphi=\omega t$ accumulate, and measures the fringe $P(0)=\tfrac12(1+\cos\omega t)$ to pin $\omega$ to fifteen or more digits. And Berry's geometric phase is now a control knob: holonomic gates in transmons and trapped ions imprint a relative phase from a control loop's geometry, giving robustness to certain timing errors. The open frontier — from diagnosing dephasing via lost fringe visibility to engineering geometric-phase-protected qubits — is the practical art of protecting relative phases while ignoring global ones.

The physical picture

Picture each amplitude as a little clock, its hand pointing along that amplitude's phase. The global phase winds every clock forward together; with no reference clock outside the system, no experiment can register this uniform advance. A relative phase does something you can catch: it winds some clocks more than others, so the hands disagree, and that disagreement shows up as interference when the branches recombine. Probabilities in the computational basis read only the lengths of the hands, $|\alpha|^2$ and $|\beta|^2$ ; interference in a rotated basis reads their relative angle, $\varphi$ . Geometrically the same story is the Bloch sphere. A normalized qubit is a unit vector in $\mathbb{C}^2$ , the three-sphere $S^3$ ; quotienting out the global phase (the Hopf fibration $S^3\to S^2$ ) collapses each phase-circle to a point and leaves the two-sphere $S^2$ of distinct physical states. The polar angle $\vartheta$ comes from the magnitudes and the azimuth $\varphi$ is the relative phase. The first figure shows where the two phases live — $|+\rangle$ and $|-\rangle$ at antipodal equator points, differing only by a relative phase of $\pi$ .

Figure

The Bloch sphere: amplitude magnitudes set the polar angle ϑ while the relative phase φ is the azimuth. The plus state sits at φ = 0 and the minus state at φ = π, antipodal and orthogonal though they differ only by a relative phase of π. A global phase e^{iθ} slides the state along the invisible Hopf fibre and never moves the point.

The second figure is where the relative phase stops being geometry and becomes a measured number. Prepare $\tfrac{1}{\sqrt2}(|0\rangle+e^{i\varphi}|1\rangle)$ , sweep $\varphi$ , and measure in the Hadamard basis. The counts trace a cosine fringe — the archetypal interference signature, and the exact shape a Ramsey sequence draws. That $\varphi=0$ (all counts to $+$ ) and $\varphi=\pi$ (all counts to $-$ ) give opposite results is the operational proof that the relative phase is physical — while the global phase never moves any count anywhere.

Figure

Interference fringe from sweeping the relative phase φ and measuring in the Hadamard basis: P(+) = ½(1 + cos φ) and P(−) = ½(1 − cos φ). At φ = 0, P(+) = 1; at φ = π, P(+) = 0; the curves cross at ½ when φ = π/2. Full-swing visibility is 1.

The core equations

Two kets differing by an overall unit-modulus factor are declared physically equivalent; the tilde denotes "same ray, same density operator, same everything you can measure."

|\psi'\rangle = e^{i\theta}\,|\psi\rangle \;\sim\; |\psi\rangle \qquad (1)

where: $|\psi\rangle$ — normalized state vector, $\langle\psi|\psi\rangle=1$ ; $\theta$ — real global phase angle; $\sim$ — physical equivalence (same ray in $\mathbb{CP}^{n-1}$ ). Reading: the whole circle of states swept by $\theta$ is a single physical point. Stripping that redundancy leaves exactly two real angles; the standard parametrization puts the relative phase where you can see it — as the azimuth $\varphi$ , with the magnitudes in a polar angle $\vartheta$ .

|\psi\rangle=\cos\tfrac{\vartheta}{2}\,|0\rangle+e^{i\varphi}\sin\tfrac{\vartheta}{2}\,|1\rangle \qquad (2)

where: $\vartheta$ — polar angle, $0\le\vartheta\le\pi$ , fixed by $|\alpha|,|\beta|$ ; $\varphi$ — relative phase and azimuthal angle, $0\le\varphi<2\pi$ . Reading: every pure qubit state is one point $(\vartheta,\varphi)$ on the Bloch sphere, with no room for a global phase. The operational content of "unobservable" is that Born probabilities do not budge under a global phase: the amplitude picks up $e^{i\theta}$ , the modulus-squared eats it, the probability is untouched.

P(\phi)=|\langle\phi|\psi'\rangle|^2=|e^{i\theta}\langle\phi|\psi\rangle|^2=|\langle\phi|\psi\rangle|^2 \qquad (3)

where: $|\phi\rangle$ — any measurement-outcome state; $P(\phi)$ — its Born probability. Reading: the same argument gives $\langle A\rangle_{\psi'}=\langle\psi|e^{-i\theta}Ae^{i\theta}|\psi\rangle=\langle\psi|A|\psi\rangle$ for every observable. The density operator has already done this quotient for you: building $\rho$ pairs a $+\theta$ from $|\psi\rangle$ with a $-\theta$ from $\langle\psi|$ , so the global phase annihilates itself, while the relative phase is stored openly in the off-diagonal coherences.

\rho=|\psi\rangle\langle\psi|=\frac12\begin{pmatrix}1+\cos\vartheta & e^{-i\varphi}\sin\vartheta\\ e^{i\varphi}\sin\vartheta & 1-\cos\vartheta\end{pmatrix} \qquad (4)

where: diagonal entries — populations $|\alpha|^2,|\beta|^2$ , blind to all phase; off-diagonal $\rho_{01}=\alpha\beta^{*}=\tfrac12 e^{-i\varphi}\sin\vartheta$ — the coherence carrying $\varphi$ . Reading: $\big(e^{i\theta}|\psi\rangle\big)\big(e^{-i\theta}\langle\psi|\big)=\rho$ is unchanged, while dephasing that erases the off-diagonals destroys $\varphi$ but not the populations. Rotate the basis and the coherence surfaces as a probability: state (2) in the Hadamard ( $X$ ) basis gives the interference law the fringe figure plots.

P(\pm)=\tfrac12\bigl(1\pm 2|\alpha||\beta|\cos\varphi\bigr)=\tfrac12\bigl(1\pm\sin\vartheta\cos\varphi\bigr) \qquad (5)

where: $P(\pm)$ — probabilities of the $X$ -eigenstates $|\pm\rangle$ ; the cross term $2|\alpha||\beta|\cos\varphi$ — the interference signature carrying $\varphi$ . Reading: the global phase is still absent, but the relative phase now modulates the counts between $0$ and $1$ — the coherence read out as a fringe. Finally, relative phases are programmed by the diagonal phase gates, Pauli $Z$ the unit generator — the gates a compiler uses to steer $\varphi$ , absorbing any overall factor for free.

Z=\begin{pmatrix}1&0\\0&-1\end{pmatrix},\quad S=\begin{pmatrix}1&0\\0&i\end{pmatrix},\quad T=\begin{pmatrix}1&0\\0&e^{i\pi/4}\end{pmatrix}:\ \varphi\to\varphi+\{\pi,\tfrac\pi2,\tfrac\pi4\} \qquad (6)

where: $Z,S,T$ — the standard diagonal gates adding $\pi,\pi/2,\pi/4$ to the relative phase. Reading: gate equivalence is judged in $PU(2)=U(2)/U(1)$ , so $S$ and $\sqrt Z$ count as the same operation despite a global-phase difference.

Derivation AWhy every qubit state reduces to two real angles

The goal is to show that a general normalized qubit, starting with four real parameters, collapses to exactly two physical angles once normalization and the global-phase redundancy are imposed — and that the surviving azimuth is precisely the relative phase. This is the concrete meaning of "pure states are rays," with the Bloch sphere as what survives the quotient. We count parameters, pass to polar form, and factor out the overall phase.

Step 1. Write the general state $|\psi\rangle=\alpha|0\rangle+\beta|1\rangle$ with $\alpha,\beta\in\mathbb{C}$ . Two complex numbers are four real numbers. Why: the raw content of a vector in $\mathbb{C}^2$ before any physical constraint.

|\psi\rangle=\alpha|0\rangle+\beta|1\rangle,\qquad \alpha,\beta\in\mathbb{C}

Step 2. Impose normalization $|\alpha|^2+|\beta|^2=1$ . One real equation removes one parameter, leaving three. Why: the constraint lets the magnitudes be written as a cosine and sine of a single angle.

|\alpha|^2+|\beta|^2=1\;\Rightarrow\;|\alpha|=\cos\tfrac{\vartheta}{2},\ |\beta|=\sin\tfrac{\vartheta}{2}

Step 3. Put the amplitudes in polar form $\alpha=|\alpha|e^{i\phi_0}$ , $\beta=|\beta|e^{i\phi_1}$ , then factor the common phase $e^{i\phi_0}$ out front. By equivalence (1) that overall factor is discardable. Why: this is the single moment where the global phase is removed; everything after it is genuinely physical.

|\psi\rangle=e^{i\phi_0}\big(\cos\tfrac{\vartheta}{2}|0\rangle+e^{i(\phi_1-\phi_0)}\sin\tfrac{\vartheta}{2}|1\rangle\big)\;\sim\;\cos\tfrac{\vartheta}{2}|0\rangle+e^{i\varphi}\sin\tfrac{\vartheta}{2}|1\rangle

Step 4. Read off the two survivors: the polar angle $\vartheta\in[0,\pi]$ from the magnitudes and the relative phase $\varphi=\phi_1-\phi_0\in[0,2\pi)$ . These are spherical coordinates of a point on $S^2$ . Why: two angles on a sphere is $\mathbb{CP}^1\cong S^2$ , the Bloch sphere; the only physical phase is the difference $\varphi$ .

(\vartheta,\varphi)\in S^2,\qquad \varphi=\arg\beta-\arg\alpha

Note: the parameter count runs $4\to3\to2$ ; at $\vartheta=\pi/2$ the state is $\tfrac{1}{\sqrt2}(|0\rangle+e^{i\varphi}|1\rangle)$ , the equator of the Bloch sphere, confirming that the only surviving phase is the relative azimuth $\varphi$ .

Go deeper · Derivation B · 1 minUnobservability of the global phase from the measurement postulate

Derivation BUnobservability of the global phase from the measurement postulate

The goal is a rigorous proof — not a slogan — that $|\psi\rangle$ and $e^{i\theta}|\psi\rangle$ give identical statistics for every conceivable measurement, and so are physically one state. Checking a single basis is not enough; we prove invariance for an arbitrary POVM, then restate it at the density-operator level where it becomes manifest. The proof rests on $e^{\pm i\theta}$ being scalars that slide freely through operators and out of inner products.

Step 1. Take a general measurement with elements $\{M_m\}$ , $M_m\succeq0$ , $\sum_m M_m=\mathbb{1}$ , and write the outcome probability for $|\psi'\rangle=e^{i\theta}|\psi\rangle$ . Why: arbitrary POVMs cover projective, noisy, and joint measurements at once.

P_m=\langle\psi'|M_m|\psi'\rangle,\qquad |\psi'\rangle=e^{i\theta}|\psi\rangle

Step 2. Substitute and pull the scalars out. The bra contributes $e^{-i\theta}$ , the ket $e^{+i\theta}$ , and being numbers they commute past $M_m$ . Why: the crux — a scalar never entangles with the operator content of a measurement.

P_m=\langle\psi|e^{-i\theta}M_m e^{i\theta}|\psi\rangle=e^{-i\theta}e^{i\theta}\langle\psi|M_m|\psi\rangle

Step 3. Cancel $e^{-i\theta}e^{i\theta}=1$ . The probability is identical to that of the unphased state, for every $m$ and every measurement, so no experiment separates them. Why: "for every measurement" upgrades this to genuine physical identity.

P_m=\langle\psi|M_m|\psi\rangle\qquad\text{for all }m

Step 4. Restate operationally: all statistics enter through $\rho$ , which is phase-blind by construction. Why: the basis-free statement — the physical state is $\rho$ , which has already discarded the phase.

\rho'=e^{i\theta}|\psi\rangle\langle\psi|e^{-i\theta}=|\psi\rangle\langle\psi|=\rho,\qquad P_m=\mathrm{Tr}(M_m\rho)

Note: normalization is preserved throughout — $\sum_m P_m=\mathrm{Tr}(\rho)=1$ holds for both $|\psi\rangle$ and $e^{i\theta}|\psi\rangle$ , as expected from the unit-trace of any density operator.

Go deeper · Derivation C · 1 minPromotion of a global phase to a relative phase by a controlled gate

Derivation CPromotion of a global phase to a relative phase by a controlled gate

The goal is to expose the most important subtlety of the subject: the label "global" is not absolute but depends on what you call the system. A phase that is provably unobservable on an isolated eigenstate becomes a fully observable relative phase of a control qubit the moment the two are wired together by a controlled gate — the mechanism, phase kickback, that lets quantum phase estimation read an eigenvalue and makes Shor's algorithm run. We prepare the target in an eigenstate, act with the controlled unitary on a superposed control, and watch the phase migrate from unobservable to observable.

Step 1. Let $|u\rangle$ be an eigenstate of a unitary $U$ with eigenvalue $e^{i\theta}$ . On $|u\rangle$ alone that eigenvalue is a pure global phase — by the previous derivation, invisible. Why: the baseline — in isolation the target's phase is unmeasurable.

U|u\rangle=e^{i\theta}|u\rangle

Step 2. Prepare the control in $|+\rangle$ and apply the controlled gate $\mathrm{c}U$ , which acts with $U$ on the target only when the control is $|1\rangle$ . Why: the control superposition is the reference interference needs to compare against.

\mathrm{c}U\,\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)|u\rangle=\tfrac{1}{\sqrt2}\big(|0\rangle|u\rangle+|1\rangle\,U|u\rangle\big)

Step 3. Substitute the eigenvalue and factor the target back out. The phase $e^{i\theta}$ detaches from the target and reattaches to the $|1\rangle$ branch of the control. Why: the "kickback" — the target is unchanged and the phase now sits on the control's branches.

=\tfrac{1}{\sqrt2}\big(|0\rangle+e^{i\theta}|1\rangle\big)|u\rangle

Step 4. Read the control: it now holds a relative phase $\theta$ , exactly the form of state (2) with $\vartheta=\pi/2$ , measurable by a Hadamard then a computational-basis measurement through the fringe (5). Why: the once-unobservable phase is now the control's azimuth — global promoted to relative by enlarging the system.

H\,\tfrac{1}{\sqrt2}(|0\rangle+e^{i\theta}|1\rangle):\quad P(0)=\tfrac12(1+\cos\theta)

Note: the result reduces correctly to the identity when $\theta=0$ (giving $P(0)=1$ ), and the readout $P(0)=\tfrac12(1+\cos\theta)$ is precisely the one-qubit phase-estimation primitive.

Limiting cases

Watch how the distinction plays out in each standard regime — which measurement, and which viewpoint, exposes or hides which phase.

Case	Result	Meaning
Computational-basis measurement	$P(0)=\lvert\alpha\rvert^2,\ P(1)=\lvert\beta\rvert^2$	Both phases invisible — $Z$ -basis counts see only magnitudes.
$X$ - or $Y$ -basis measurement	$P(\pm_X)=\tfrac12(1\pm\sin\vartheta\cos\varphi)$ , $P(\pm_Y)=\tfrac12(1\pm\sin\vartheta\sin\varphi)$	Relative phase $\varphi$ fully observable as a fringe; global phase still cancels.
Density-operator view	$e^{i\theta}\lvert\psi\rangle\Rightarrow\rho=\lvert\psi\rangle\langle\psi\|$ unchanged	Global phase auto-discarded; coherences $\rho_{01}=\alpha\beta^{*}$ store $\varphi$ .
Controlled operation	$\lvert0\rangle\lvert u\rangle+e^{i\theta}\lvert1\rangle\lvert u\rangle$	A would-be global phase becomes a relative phase of the joint state — phase kickback.
Gate compilation	$U\sim e^{i\alpha}U$ in $PU(2)=U(2)/U(1)$	Overall gate phases are free to discard — until the gate is used as a controlled operation.

Common trapThe belief that phases never matter, because the global phase is unobservable

The unobservability of the global phase is often over-generalized into "phases don't matter" — which is exactly backwards. Only the single overall factor common to the whole state is unobservable. Every relative phase is fully physical: it is the azimuth on the Bloch sphere, it lives in the coherences of $\rho$ , and it is the whole content of interference. Worse, whether a phase counts as "global" is contextual — the eigenvalue phase of a unitary is global on its eigenvector but becomes observable the instant that eigenvector is the target of a controlled gate, the operating principle of controlled- $Z$ , phase kickback, and phase estimation. Discard an overall factor freely; never discard a phase difference, and never assume a phase stays global when its system is enlarged.

Worked examples

1 · Reading a relative phase as an interference fringe (fully worked)

A qubit $|\psi\rangle=\tfrac{1}{\sqrt2}(|0\rangle+e^{i\varphi}|1\rangle)$ is measured in the Hadamard basis. Compute $P(+)$ at $\varphi=0,\ \tfrac{\pi}{2},\ \pi$ , find the visibility, and confirm a global phase changes nothing.

Step 1. Overlap with $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ . Using $\langle+|0\rangle=\langle+|1\rangle=\tfrac{1}{\sqrt2}$ ,

\langle+|\psi\rangle=\tfrac{1}{\sqrt2}\cdot\tfrac{1}{\sqrt2}(1+e^{i\varphi})=\tfrac12(1+e^{i\varphi}).

Step 2. Square the modulus, using $|1+e^{i\varphi}|^2=(1+\cos\varphi)^2+\sin^2\varphi=2+2\cos\varphi$ :

P(+)=\big|\tfrac12(1+e^{i\varphi})\big|^2=\tfrac14(2+2\cos\varphi)=\tfrac12(1+\cos\varphi).

Step 3. Evaluate at the three phases. At $\varphi=0$ : $P(+)=\tfrac12(1+1)=1$ — the state is $|+\rangle$ , perfect constructive interference. At $\varphi=\tfrac{\pi}{2}$ : $P(+)=\tfrac12(1+0)=0.5$ — the state is $|{+}i\rangle$ , and the $X$ basis is maximally uninformative. At $\varphi=\pi$ : $P(+)=\tfrac12(1-1)=0$ — the state is $|-\rangle$ , perfect destructive interference.

Step 4. Visibility $V=\dfrac{P_{\max}-P_{\min}}{P_{\max}+P_{\min}}=\dfrac{1-0}{1+0}=\mathbf{1}$ : a full-contrast fringe. Now attach a global phase, $|\psi\rangle\to e^{i\theta}|\psi\rangle$ : then $P(+)\to|e^{i\theta}|^2\,\tfrac12(1+\cos\varphi)=\tfrac12(1+\cos\varphi)$ — identical. The fringe depends only on $\varphi$ ; the global phase $\theta$ never appears. Result: $P(+)=1,\ 0.5,\ 0$ at $\varphi=0,\tfrac\pi2,\pi$ ; visibility $V=1$ ; global phase irrelevant.

2 · Distinguishing global from relative, with the density operator as referee

Decide whether each pair is the same physical state: (a) $|+\rangle$ and $e^{i\pi/3}|+\rangle$ ; (b) $|+\rangle$ and $|-\rangle$ . Then for $|\psi\rangle=\cos\tfrac{\vartheta}{2}|0\rangle+e^{i\varphi}\sin\tfrac{\vartheta}{2}|1\rangle$ with $\vartheta=\tfrac{\pi}{2},\ \varphi=\tfrac{\pi}{3}$ , compute $\rho$ , $\langle Z\rangle$ , and $\langle X\rangle$ .

Step 1. Pair (a): the two kets differ by the overall factor $e^{i\pi/3}$ , a global phase. Their density operators coincide, $e^{i\pi/3}|+\rangle\langle+|e^{-i\pi/3}=|+\rangle\langle+|$ , so they are the same physical state.

Step 2. Pair (b): $|+\rangle$ and $|-\rangle$ differ by a relative phase ( $\varphi=0$ vs $\varphi=\pi$ ), not a global one. They are orthogonal, $\langle+|-\rangle=0$ , hence perfectly distinguishable by an $X$ -basis measurement — different physical states.

Step 3. Build $\rho$ from (4) with $\vartheta=\tfrac{\pi}{2}$ (so $\cos\vartheta=0$ , $\sin\vartheta=1$ ) and $\varphi=\tfrac{\pi}{3}$ (so $\cos\varphi=\tfrac12,\ \sin\varphi=\tfrac{\sqrt3}{2}$ ):

\rho=\frac12\begin{pmatrix}1 & e^{-i\pi/3}\\ e^{i\pi/3} & 1\end{pmatrix}=\begin{pmatrix}0.5 & 0.25-0.433\,i\\ 0.25+0.433\,i & 0.5\end{pmatrix}.

Step 4. Expectation values from $\langle A\rangle=\mathrm{Tr}(A\rho)$ . $\langle Z\rangle=\rho_{00}-\rho_{11}=0.5-0.5=0$ — phase-blind, reading only the equal populations. $\langle X\rangle=2\,\mathrm{Re}\,\rho_{01}=\cos\varphi=0.5$ and $\langle Y\rangle=-2\,\mathrm{Im}\,\rho_{01}=\sin\varphi\approx0.866$ . The relative phase is entirely in the transverse components $(\langle X\rangle,\langle Y\rangle)=(\cos\varphi,\sin\varphi)$ — the Bloch vector's azimuth is $\varphi$ , exactly as the geometry promised. Result: (a) same state, (b) different; $\langle Z\rangle=0$ , $\langle X\rangle=0.5$ , $\langle Y\rangle\approx0.866$ .

Physics of the portrait

The distinction looks like bookkeeping until its deepest consequence: a state carried around a loop can return to the same physical point — the same ray — yet emerge carrying an extra, completely observable phase, set not by dynamics but by the geometry of the path. Its modern discoverer gives this article its portrait.

Origins · 1 minMichael Berry and how this was discovered

Michael Berry

b. 1941 · geometric phase and semiclassical physics

Michael Berry, a mathematical physicist at Bristol, asked in 1984 a question that seemed too simple to be new: if a quantum system is guided slowly around a closed loop in its parameter space, returning exactly to its starting state, what phase has it accumulated? Beyond the expected dynamical phase $-\tfrac{1}{\hbar}\int E\,\mathrm{d}t$ , Berry found a second, purely geometric contribution depending only on the shape of the loop, not on how fast it is traversed. It is the cleanest illustration of this article's theme: the state comes home to the same ray (globally the same physical point), yet the accumulated phase is observable because it is relative to a reference beam, exactly as in the fringe (5). Berry showed it equals the flux of a curvature through the loop — mathematically a magnetic field in parameter space. The idea had deep roots Berry generously traced. S. Pancharatnam had, in 1956, defined the relative phase between two states of polarized light and found it geometric — now recognized as the Berry phase for the qubit of polarization. The Aharonov–Bohm effect of 1959 was another instance: a charged particle encircling a confined magnetic flux it never touches picks up an observable relative phase set purely by the enclosed flux. Barry Simon recognized Berry's curvature as a fibre-bundle holonomy, and Aharonov and Anandan soon removed the slow-evolution assumption. Within two years Tomita and Chiao measured the phase in a coiled optical fibre; today it is engineered into holonomic quantum gates. Berry turned a subtlety about "unobservable" phases into a resource.

History

1926

— Born's rule fixes every prediction as |amplitude|², making an overall phase unmeasurable in principle

1956

— Pancharatnam defines the relative phase between two polarization states and finds it is geometric

1959

— Aharonov and Bohm predict an observable phase from a magnetic flux the charge never touches

1984

— Berry shows an adiabatic cyclic evolution imprints a geometric phase set only by the loop's shape

1987

— Aharonov and Anandan free the geometric phase from the adiabatic assumption

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, §1.3, §2.2.7, §5.2 · Preskill, Lecture Notes on Quantum Computation, Ch. 2 · Watrous, The Theory of Quantum Information, Ch. 1–2 · Wilde, Quantum Information Theory, Ch. 3 · Kitaev, Shen & Vyalyi, Classical and Quantum Computation, Ch. 8 · Mermin, Quantum Computer Science, Ch. 1. For the geometric phase: Berry (Proc. R. Soc. A, 1984), Pancharatnam (1956), and Shapere & Wilczek, Geometric Phases in Physics. The key results collected in this article are: (1) the parameter count $4\to3\to2$ reducing a qubit to $(\vartheta,\varphi)$ on $S^2$ ; (2) the cancellation $\langle\psi|e^{-i\theta}M_m e^{i\theta}|\psi\rangle=\langle\psi|M_m|\psi\rangle$ and the density-operator invariance $e^{i\theta}\rho e^{-i\theta}=\rho$ ; (3) the interference law $P(+)=\tfrac12(1+\cos\varphi)$ from $|\tfrac12(1+e^{i\varphi})|^2$ , giving visibility $1$ ; (4) the density matrix (4) and its Bloch components $\langle X\rangle=\cos\varphi$ , $\langle Y\rangle=\sin\varphi$ , $\langle Z\rangle=\cos\vartheta$ ; (5) phase kickback $\mathrm{c}U(|+\rangle|u\rangle)=\tfrac{1}{\sqrt2}(|0\rangle+e^{i\theta}|1\rangle)|u\rangle$ . The phase increments $\pi,\pi/2,\pi/4$ for $Z,S,T$ are standard values.

Prerequisites. The Qubit and Its State Space · Dirac Bra-Ket Notation · Inner Product, Norm, and Normalization · Superposition Principle · Born Rule. Leads to. Density Operators · Pure versus Mixed States · Pauli Operators · Projective Measurements.

Ask the tutor: Ask me to walk a state around the Bloch sphere and show which measurement turns its hidden relative phase into a fringe.

If the global phase is truly unobservable, why carry it in the math at all?

Because it becomes observable the moment the system is enlarged. On an isolated state the factor $e^{i\theta}$ cancels in every prediction, so you could drop it — and compilers of single-qubit gates do. But the instant that state is one branch of a bigger superposition, say the target of a controlled gate, its "global" phase is promoted to a relative phase of the composite and starts modulating real fringes. Keeping it in the bookkeeping is what lets phase kickback and phase estimation come out right. The rule: a global phase is discardable only for the closed system it is global on, never for a subsystem you might later entangle.

What is the precise difference between a ray and a vector here?

A vector is a specific element $|\psi\rangle$ ; a ray is the whole set $\{c|\psi\rangle:c\neq0\}$ of nonzero scalar multiples, or for normalized states the circle $\{e^{i\theta}|\psi\rangle\}$ . Physical pure states are rays, not vectors, which is why the true state space is the projective Hilbert space $\mathbb{CP}^{n-1}$ rather than $\mathbb{C}^n$ . For one qubit that is $\mathbb{CP}^1$ , geometrically the two-sphere $S^2$ — the Bloch sphere. The density operator $\rho=|\psi\rangle\langle\psi|$ is the coordinate-free name for a ray, being identical for every vector in it.

Does the relative phase depend on which basis I choose?

Yes — "relative phase" is always relative to a chosen decomposition. The $\varphi$ in $\alpha|0\rangle+e^{i\varphi}|\beta||1\rangle$ is the phase of $\beta$ relative to $\alpha$ in the computational basis; rewrite the same state in the $X$ basis and the coefficients carry a different relative phase. What is basis-independent is the physical state — the point on the Bloch sphere. Choosing a measurement basis chooses which Bloch component you read: the $Z$ basis reads $\cos\vartheta$ and is blind to $\varphi$ , while $X$ and $Y$ read $\sin\vartheta\cos\varphi$ and $\sin\vartheta\sin\varphi$ and so expose it. The global phase is unobservable in every basis at once.

How does decoherence connect to the relative phase?

Dephasing is precisely the loss of relative-phase information. In the density matrix (4) the relative phase lives in the coherences $\rho_{01}=\alpha\beta^{*}$ ; dephasing shrinks those toward zero while leaving the populations intact. The definite relative phase blurs into a random one, and a coherent superposition decays into a classical mixture. You detect it as lost interference visibility — example 1's fringe fades to a flat line. Distinguishing a definite-but-unknown relative phase from a genuinely phase-averaged mixture is exactly the pure-versus-mixed question, diagnosed by the purity $\mathrm{Tr}(\rho^2)$ or by visibility.

Why can two different gates like S and √Z be treated as the same?

Because they differ only by a global phase, and gate equivalence for state transformation is judged in $PU(2)=U(2)/U(1)$ . Here $\sqrt Z=\mathrm{diag}(1,i)=S$ exactly, but generally any $U$ and $e^{i\alpha}U$ implement the identical Bloch-sphere rotation and are interchangeable on a lone qubit — a freedom compilers exploit constantly. The caveat is this article's recurring theme: it evaporates when the gate becomes a controlled operation, because then $e^{i\alpha}$ becomes a physical relative phase of the control and can no longer be dropped.

Is the geometric (Berry) phase a global or a relative phase?

Both, depending on your vantage — which is why it is such a clean illustration. After a cyclic evolution the state returns to the same ray, so relative to itself the geometric phase is global and changes none of the system's own statistics. But relative to a reference — a second beam or a control branch that did not traverse the loop — it is an observable relative phase, and that is how every geometric-phase experiment reads it out. So it is unobservable if you look only at the system that acquired it, observable the instant you compare against something that did not: contextual promotion of a global phase to a relative one, made by geometry instead of a controlled gate.