Global vs Relative Phase

Here is a puzzle that stops most students the first time they meet it. You have the state

the simplest qubit state in the world. Now multiply it by e^{i\pi/3}:

Is |\psi'\rangle a new state? It looks different on paper. The amplitude has changed — it used to be 1 and now it is a complex number with a non-trivial phase. Surely something has changed?

The answer is no. |\psi'\rangle is exactly the same physical state as |\psi\rangle. No experiment, no measurement, no observation of any kind can distinguish them. They sit at exactly the same point on the Bloch sphere. They behave identically under every subsequent quantum operation that acts on them alone. For all physical purposes they are one state with two notations.

Now change the setup slightly. Start with

and multiply only the |1\rangle component by e^{i\pi/3}:

Is |\phi'\rangle a new state? This time, yes. |\phi'\rangle sits at a different point on the Bloch sphere from |\phi\rangle. A measurement in the X-basis gives different probabilities. The two states are physically distinct in every sense.

Same operation in both cases — multiplication by e^{i\pi/3}. First time, no change. Second time, a real change. What is different?

The first was a global phase — the whole state got multiplied, and the phase cancelled out of every observable quantity. The second was a relative phase — only the |1\rangle component got multiplied, and the resulting shift between the |0\rangle and |1\rangle amplitudes is real and measurable.

This distinction is one of the two or three most confusing points in all of quantum computing — and unlike the "0 and 1 at the same time" misconception, this one trips up people who actually understand quantum mechanics at a working level. Pop-science does not even mention it; formal textbooks state it once and move on; but if you get it wrong in a circuit design, everything downstream breaks.

This chapter builds it carefully. First the rule (global phase is invisible, relative phase is visible), then the proof (from the probability formula), then the Bloch-sphere picture (why both can be read off the sphere), then — the subtle part — when a global phase starts as a global phase but becomes a relative phase inside a controlled gate. That last move is how half of quantum computing works.

The rule — stated once, then unpacked

Global phase. If you multiply an entire quantum state |\psi\rangle by a complex number e^{i\varphi} with |e^{i\varphi}| = 1, you get a new symbolic expression e^{i\varphi}|\psi\rangle that represents the same physical state as |\psi\rangle. Every physically measurable prediction — probabilities, expectation values, correlation functions — is identical.

Relative phase. If you multiply just one component of a superposition by e^{i\varphi} — say, you change \alpha|0\rangle + \beta|1\rangle into \alpha|0\rangle + e^{i\varphi}\beta|1\rangle — you have made a physically distinct state. Different measurement probabilities in some bases, different position on the Bloch sphere, different behaviour under subsequent gates.

In short: phases on the whole state are nothing; phases between components are everything.

Proof that global phase is invisible

The claim is strong — every observable is identical — so it deserves an actual proof.

Start with the Born rule. Given a state |\psi\rangle, the probability of obtaining outcome |k\rangle in a projective measurement is

Now replace |\psi\rangle with e^{i\varphi}|\psi\rangle. The inner product becomes

Take the modulus squared:

Why the phase drops out: |e^{i\varphi}|^2 = \cos^2\varphi + \sin^2\varphi = 1 identically for any real \varphi. The modulus squared is blind to the phase.

So every probability is identical. Since everything you can measure in quantum mechanics reduces to a sequence of probabilities (expectation values are just weighted averages of them), every observable is identical. |\psi\rangle and e^{i\varphi}|\psi\rangle are indistinguishable by any physical test.

This is not a special property of the computational basis; the same argument goes through for any measurement basis \{|k\rangle\}. It is not a special property of projective measurements; the same argument goes through for any positive-operator-valued measure. Phase on the whole state, no effect on anything.

Why relative phase is visible

Consider |+\rangle and a modified version:

Measure each in the X-basis. The projector onto |+\rangle is |+\rangle\langle+|. For the first state, the overlap is

For the modified state, compute

Modulus squared:

Why this calculation matters: the probability of measuring |+\rangle went from 1 (for the original |+\rangle) to \tfrac{1}{2}(1+\cos\varphi) for the phase-shifted state. If \varphi = \pi, the probability drops to 0: the state has become |-\rangle, orthogonal to |+\rangle. Relative phase is visible — you can literally read it off a measurement statistic.

So when \varphi = 0, p_+ = 1; when \varphi = \pi, p_+ = 0; and for intermediate values the probability varies smoothly. The relative phase shows up as a detectable difference in measurement statistics.

Contrast this with the global phase. Multiplying |+\rangle by e^{i\varphi} gives

The amplitude of |0\rangle and the amplitude of |1\rangle both picked up the same e^{i\varphi}. The probability of measuring |+\rangle is |\langle+|e^{i\varphi}|+\rangle|^2 = |e^{i\varphi}|^2 \cdot 1 = 1 — unchanged. Same e^{i\varphi}, completely different effect.

The amplitude-ratio view — the cleanest framing

Here is the way to think about global vs relative phase that makes the distinction obvious and never trips you up again.

Given a single-qubit state |\psi\rangle = \alpha|0\rangle + \beta|1\rangle, the "shape" of the state — the thing that distinguishes it from any other state — is encoded in the ratio \beta / \alpha.

This is a complex number. Its modulus tells you how the probability is split between |0\rangle and |1\rangle. Its phase tells you the relative phase between the two components. Together, those two pieces of information — a magnitude and a phase, or equivalently two real numbers — encode exactly the physical state.

Now multiply both \alpha and \beta by the same phase factor e^{i\varphi}:

The ratio is unchanged. So the physical state is unchanged. Global phase is a phase factor that multiplies both the numerator and the denominator of \beta/\alpha, and therefore changes nothing.

Now multiply only \beta by e^{i\varphi}:

The ratio picks up the phase. The physical state has changed. Relative phase is a phase change in the ratio.

The Bloch-sphere connection

The Bloch sphere has already quietly encoded the global-vs-relative distinction. A general single-qubit state is parameterised as

Notice: only two angles. There is no slot for an extra "global phase" parameter. The Bloch sphere is built to ignore global phase by construction — the space of single-qubit states modulo global phase is exactly the 2-sphere.

Why there is no global-phase slot: mathematically, quantum states live not in a vector space but in the projective Hilbert space — the set of unit vectors modulo global phase. The Bloch sphere is a concrete realisation of this quotient for a single qubit. Two vectors that differ by a global phase are, in the projective sense, literally the same point.

Now look at the two angles:

• \theta is the polar angle (latitude): how much of the state is |0\rangle vs |1\rangle. This corresponds to the magnitudes of \alpha and \beta.
• \varphi is the azimuthal angle (longitude): the relative phase between |0\rangle and |1\rangle.

So on the Bloch sphere: changing \theta changes the probability balance; changing \varphi changes the relative phase. A state at the |+\rangle position has \varphi = 0. A state at |-\rangle has \varphi = \pi. A state at |+i\rangle has \varphi = \pi/2. All six cardinal equatorial states are identical in magnitude structure (|+\rangle, |-\rangle, |+i\rangle, |-i\rangle all have \theta = \pi/2) — they differ only in relative phase.

Two key consequences:

1. A global phase moves the state through the projective space's abstract geometry, but the Bloch point doesn't budge. The two angles (\theta, \varphi) don't change because global phase doesn't touch them.

2. A relative phase between |0\rangle and |1\rangle is exactly what \varphi is. Change the relative phase by \Delta\varphi and the Bloch point spins around the equator by the same \Delta\varphi.

The Bloch sphere is, in a precise sense, the "observable state space" of a single qubit. Global phase is the thing you have to quotient out to get there.

Worked example 1 — two states, one Bloch point

Let's actually do the computation on a specific pair of states.

When a global phase stops being global — the controlled-gate twist

Now the subtle move. Start with a single-qubit gate U and a gate U' = e^{i\varphi} U that differs from U only by a global phase. Acting alone on a qubit, they do the same thing — same rotations, same measurements, same everything. No experiment can tell them apart.

But embed them inside a controlled operation. Build the controlled-U gate: "if the control qubit is |1\rangle, apply U to the target; if the control is |0\rangle, do nothing." Now do the same for controlled-U'. Are these two two-qubit gates the same?

Write out the two-qubit matrices. Let |c\rangle be the control and |t\rangle the target. Controlled-U acts as:

Controlled-U', where U' = e^{i\varphi} U:

On the |0\rangle|t\rangle branch, the two gates are identical. But on the |1\rangle|t\rangle branch, controlled-U' picks up an extra e^{i\varphi} that controlled-U does not.

So on a superposition state — say \tfrac{1}{\sqrt{2}}(|0\rangle + |1\rangle)|t\rangle — controlled-U gives

while controlled-U' gives

The e^{i\varphi} is now sitting between the |0\rangle and |1\rangle components of the control qubit. It has become a relative phase — a phase difference between the two branches of the superposition on the control. And relative phases are observable.

Why this happens: inside the larger two-qubit state, the "whole" is not just U's output on the target — it is a sum of two branches (controlled on |0\rangle and |1\rangle), each with its own target output. The global phase attached to U only multiplies one branch, not both, so it becomes relative.

This is not a technicality. It is the mechanism behind phase kickback, the trick at the heart of Deutsch's algorithm, Deutsch-Jozsa, quantum phase estimation, and Shor's factoring algorithm. Kickback works by arranging for a "global" phase attached to a gate U on an eigenstate of U to become a relative phase on the control — where a Hadamard at the end can read it out. Without the global-becomes-relative distinction, none of this works.

Worked example 2 — phase kickback through controlled-Z on |+⟩|+⟩

To see the global-becomes-relative principle in action, compute the action of a controlled-Z gate on the two-qubit state |+\rangle|+\rangle.

Common confusions

This is the section that exists because this is the most error-prone topic in all of quantum computing. Read it slowly.

• "-|0\rangle = |0\rangle as a quantum state, so the minus sign never matters." Half-right. As an isolated state: yes, -|0\rangle is the same Bloch point as |0\rangle, produces the same measurement statistics, behaves identically under any subsequent single-qubit gate. But the moment -|0\rangle appears as a coefficient in a superposition — like |+\rangle vs |-\rangle — the sign becomes a relative phase and becomes fully observable. The rule is context-sensitive: global in isolation, can become relative in superpositions or under controlled gates.

• "Global phase never matters." False. It matters in three specific contexts:

  1. In a superposition: if you split a state along several paths that each pick up different "global" phases, the differences become relative phases and interfere.
  2. In a controlled gate: a global phase on U becomes a relative phase of the compound circuit.
  3. In derivations and proofs: even when the final state is the same, tracking global phases carefully is often necessary to show two circuits are equivalent, and can surface bugs in decompositions.

• "Relative phase is just the minus sign in |-\rangle." More generally, a relative phase is any complex phase e^{i\varphi} between the |0\rangle and |1\rangle components. |+i\rangle = \tfrac{1}{\sqrt{2}}(|0\rangle + i|1\rangle) has a relative phase of i = e^{i\pi/2}. |+\rangle has relative phase 1 = e^{i \cdot 0}. |-\rangle has relative phase -1 = e^{i\pi}. All six cardinal equatorial states are distinguished only by their relative phase.

• "Quantum states are vectors, and vectors with different coefficients are different vectors." Mathematically, yes, in a Hilbert space. Physically, quantum states are rays — equivalence classes of vectors that differ by a nonzero complex scalar. The projective Hilbert space is what matters physically. This is why you see the phrase "up to a global phase" everywhere in quantum literature: it is stating the equivalence relation.

• "A \pi phase flip is the same as multiplying by -1 — trivial." It is trivial as a global phase, but Z as a gate is not trivial because Z applies the factor only to the |1\rangle component, not to the whole state. Z|0\rangle = |0\rangle but Z|1\rangle = -|1\rangle. So Z creates a relative phase between |0\rangle and |1\rangle in a superposition — not a global phase. The act of "multiplying the full state by -1" would be a global phase; the act of "multiplying only the |1\rangle component by -1" is a relative phase and is exactly what Z does.

• "I can always pick a gauge where global phase is zero." True on isolated states, but once you commit to a gauge choice on each subcircuit, composing them correctly requires tracking whether any subcircuit introduces its own global phase that becomes relative at a later join. This is exactly what compilers do when they propagate phases through a circuit.

Going deeper

Projective Hilbert space — the proper home of quantum states

In a linear-algebra course you learn that the state of a qubit is a unit vector in \mathbb{C}^2 — the 2-dimensional complex vector space. Strictly speaking, this is wrong. The state of a qubit is a ray in \mathbb{C}^2 — an equivalence class of unit vectors under the relation "|\psi\rangle \sim |\psi'\rangle iff |\psi'\rangle = e^{i\varphi}|\psi\rangle for some real \varphi."

The set of rays is called the projective Hilbert space, written \mathbb{CP}^1 for a single qubit. For a qubit, \mathbb{CP}^1 is the Riemann sphere, which is topologically equivalent to the Bloch sphere. For n qubits, the projective Hilbert space is \mathbb{CP}^{2^n - 1}, a (2^{n+1} - 2)-real-dimensional complex manifold.

Why bother with this abstraction? Because it makes the global-phase-is-invisible rule automatic. You don't have to keep remembering to quotient out; the quotient is baked into the state space. And for multi-qubit systems, the projective structure is essential for correctly defining when two states are "the same" and when two circuits compute "the same" unitary up to irrelevant phase.

One physical consequence: a quantum state is not a vector but an equivalence class, so the "phase" of a state is genuinely not a physical quantity. Only differences of phases (relative phases between components of a superposition, or phases acquired over paths through parameter space) are physical.

The Berry phase — global phase that isn't quite global

Here is a subtlety: sometimes you can pick up a global phase that, despite being global, has physical consequences.

If you slowly vary the parameters of a quantum system — say, a magnetic field rotating slowly around a spin — the state adiabatically tracks an instantaneous eigenstate. At the end of a loop in parameter space (the magnetic field returns to its original orientation), the state picks up a phase. Part of it is the expected dynamical phase e^{-iEt/\hbar}. But there is an extra piece called the Berry phase (or geometric phase) that depends on the geometry of the loop in parameter space, not on how fast you traversed it.

This extra phase is mathematically a global phase — it multiplies the state that comes back from the loop. In isolation, it would be invisible. But if you split a wave packet into two paths, take one path around the loop and the other not, and then recombine, the Berry phase between the two paths becomes a relative phase, and it interferes observably. Berry-phase interference has been directly measured in neutron interferometry and in atom-optics experiments.

Berry's key insight (1984): global phase becomes observable in a topological sense — it tracks the topology of loops in the parameter space of the Hamiltonian. It is the quantum-mechanical analog of parallel transport in differential geometry, and it appears in the integer quantum Hall effect, topological insulators, and holonomic quantum computation.

The Berry phase is the cleanest example of the principle at the heart of this chapter: a quantity that looks like a global phase in one setting becomes a relative phase in another, and the apparently "unphysical" becomes physical through context.

Phase kickback — the formal statement

Let U be a unitary with an eigenstate |u\rangle and eigenvalue e^{i\lambda}:

Build a controlled-U gate with a control qubit in the state |+\rangle = \tfrac{1}{\sqrt{2}}(|0\rangle + |1\rangle) and a target in |u\rangle. Apply it:

Factor out |u\rangle:

The eigenvalue e^{i\lambda}, which was a global phase on the target |u\rangle in isolation (since U|u\rangle = e^{i\lambda}|u\rangle and |u\rangle and e^{i\lambda}|u\rangle are the same state), has been kicked onto the control, where it sits as a relative phase between the |0\rangle and |1\rangle components.

This is the central technique of quantum phase estimation: stack many controlled-U gates with different powers of U, each kicking its own power of e^{i\lambda} onto a different control qubit, then inverse-Fourier-transform the controls to read out \lambda in binary. Shor's factoring algorithm uses this exact structure — the "eigenvalue" there is the phase of a modular multiplication — and Grover's amplitude amplification has its own controlled-U moment.

Without the global-to-relative transition, phase estimation simply would not work: the information about the eigenvalue would be trapped in a physically unmeasurable global phase of the target, and no sequence of measurements could extract it.

Fault-tolerance and the cheapness of Z

In fault-tolerant quantum architectures (based on topological codes like the surface code), different gates have vastly different costs. The general trade-off:

• Clifford gates (H, S, CNOT, and Z) are "cheap" — they can be implemented transversally on the surface code, meaning a logical gate is realised by applying the same physical gate to each physical qubit that encodes the logical qubit. Error rates propagate benignly.

• Non-Clifford gates (most notably T) are "expensive" — they cannot be done transversally on the surface code. They require a separate resource called a magic state, distilled from many noisy ancilla qubits via a procedure called magic-state distillation, which itself uses many physical qubits and takes substantial time.

Where does phase fit in? Pauli Z is a Clifford gate — on the surface code it is transversal and essentially free. More remarkably, on many physical hardware platforms (including superconducting qubits, as mentioned in the previous chapter), the Z gate is implemented "virtually" by updating the phase of all subsequent pulses. No physical operation happens; a classical computer just remembers that the qubit is now -|1\rangle instead of |1\rangle.

This is possible precisely because the -1 on |1\rangle is a global phase when acting on a single basis state. The hardware doesn't need to physically flip the sign — it just needs to track the flip in software, and compensate when the next gate needs to interact with the qubit. By the time the qubit enters a superposition or a controlled gate, the tracked phase is cashed out as an actual relative phase in the circuit.

This is a beautiful piece of engineering: a gate that is "invisible" when it acts alone costs nothing to apply, and only costs something when it creates a physically observable relative phase. The compiler hides the cost inside the gates that actually see the difference.

The Raman effect — where relative phase encodes physics

One beautiful Indian example of phase carrying real physical information: the Raman effect, discovered by C. V. Raman in 1928 (Nobel Prize, 1930).

When a monochromatic light beam scatters off a molecule, most photons come out at the same frequency (elastic scattering). But a small fraction come out at a slightly different frequency, shifted by the vibrational energy of the molecule. This is the Raman scattering.

The crucial point for this chapter: the information about the molecular vibration is encoded in a relative phase between the incoming and outgoing photons. Specifically, the outgoing photon's wave function picks up a phase that depends on the instantaneous vibrational coordinate of the molecule; averaging over the molecular state gives an intensity spectrum with peaks at the shifted frequency. Without the relative-phase machinery, the Raman effect could not exist.

Modern Raman spectroscopy, widely used in Indian pharma for molecular fingerprinting and in astrophysics for exoplanet atmosphere analysis, is fundamentally a quantum-mechanical relative-phase measurement. Raman himself did not have the modern language of "relative phase," but the phenomenon he observed is exactly what happens when a quantum system's global phase gets lifted into a relative phase via interaction — the same principle that underpins quantum phase estimation.

Where this leads next

• Phase Kickback — the full chapter on the global-to-relative transition inside controlled gates.
• Quantum Fourier Transform — how nested phase kickbacks build the QFT, the engine behind Shor's algorithm.
• Phase Gates S and T — the \pi/2 and \pi/4 relative-phase gates and their role in universal computation.
• Bloch Sphere — the geometric picture that automatically quotients out global phase.
• Measurement in Arbitrary Bases — how phase structure in the state space is accessed through different measurement bases.

References

1. Wikipedia, Quantum state — includes discussion of the equivalence of rays and the role of global phase.
2. John Preskill, Lecture Notes on Quantum Computation, Ch. 2 (Foundations) — theory.caltech.edu/~preskill/ph229.
3. Nielsen and Chuang, Quantum Computation and Quantum Information, §1.2 and §2.2 (states, density operators, and the physical equivalence relation) — Cambridge University Press.
4. Qiskit Documentation, Global phase of a circuit — the practical software interface for tracking global phase in compiled circuits.
5. Wikipedia, Phase kickback — where phase estimation and phase kickback are explained together.
6. Wikipedia, Berry phase / Geometric phase — the topologically-nontrivial case of a "global" phase that becomes observable.