Measurement

Q1What is it?

Measurement is where a quantum computation finally answers a question. A qubit sits in its state,

and the state is amplitudes, not a value. Ask it “0 or 1?” — a measurement in the computational basis {|0⟩, |1⟩} — and the Born rule gives the odds:

That is the bridge the whole field crosses: quantum state ⟶ classical outcome. But measurement is more than reading a value off a register, in two ways. The outcome is only probabilistic — the state fixes the odds, not the answer. And the measurement changes the state. If the outcome is 0, the state is now |0⟩; if 1, it is |1⟩. Ask again in the same basis and you get the same answer with certainty — the state was updated to match its own readout. This is what gets called collapse, and it makes measurement an action, not an observation:

“Measurement just reveals the value the qubit already had” — the coin-flip model again, and it fails here for the same reason it failed the two-Hadamard test: before the measurement there was no value, only amplitudes. The value comes into existence with the outcome.

Q2Why does it exist?

Why a whole node for “reading out the answer”? Because two facts, easy to state and slow to absorb, sit here — and most of quantum information lives downstream of them.

First: the probabilities are not the state. Compare

Measured in the computational basis, both give P(0) = P(1) = ½ — identical statistics, indistinguishable outcomes. Yet they are different states, separated by a relative phase, and the difference is physically real: it is exactly what the second Hadamard on the Qubits & States page converts into opposite answers. One basis cannot see everything a state is.

Second: the basis is a choice. Measurement is not the fixed question “is it 0 or 1?” — it is a question you pick. Measure |+⟩ in the computational basis and you get a fair coin. Measure the same state in the X basis {|+⟩, |−⟩} and you get + with certainty, every time. Same state; definite in one basis, random in another. So the productive question is never “what will the qubit say?” but “what am I asking, and in which basis?”:

(The two-Hadamard experiment is this in disguise: an H followed by a computational-basis measurement is an X-basis measurement.)

Put the two facts together and you get the constraint that shapes all of quantum algorithm design: a state carries amplitudes you cannot simply look at. One measurement yields one outcome and disturbs the rest. Advantage, when it exists, comes from arranging interference before the readout so that the one answer you get is worth having.

Q3How does it connect?

Upstream, measurement is where the two mathematical inheritances of the qubit finally meet: linear algebra supplies the operators, probability theory receives the outcomes. The meeting point is the projector. For the computational basis, define

and the Born rule becomes a statement about operators:

This looks like notation for its own sake until you see what it buys: the operator form is the one that scales — to several qubits, to mixed states, and past projectors entirely. Downstream sits a hierarchy I am still climbing. Projective measurement is the clean case above. The general case is the POVM — positive operator-valued measure — where each outcome i gets a positive operator Eᵢ and only two conditions survive:

The operators no longer need to be orthogonal projectors, which is precisely what makes POVMs the right language for imperfect and clever measurements alike: distinguishing non-orthogonal states, tomography, noisy readout on real hardware. Sideways, the neighbors: the qubit supplies the state being asked; superposition sets the amplitudes the outcome is drawn from; phase hides from one basis and shows up in another; and gates are what you apply before measuring to change what the measurement can see.

Q4Where is it used?

Every quantum computation ends in one. The conceptual shape of the whole pipeline is:

The quantum part of the machine manipulates amplitudes; measurement is the only door those amplitudes can leave through, and it is a narrow door — one shot, one outcome, state disturbed. Three places where that narrowness turns from nuisance into structure. In error correction, syndrome measurements are engineered to ask about the noise without asking about the data — collapse aimed carefully, so the encoded state survives its own diagnosis. In quantum cryptography, BB84 turns collapse into a security guarantee: an eavesdropper must measure, a measurement in the wrong basis disturbs the state, and the disturbance is detectable — the “flaw” of measurement doing the protocol's work. And on real hardware, readout is a physical process with its own error rates — often worse than the gate errors — which is why measurement fidelity sits on every spec sheet and why characterizing readout error is its own discipline.

To make it concrete: in Qiskit, prepare |+⟩ and measure — a fair coin, run after run. Now insert one H before the measurement and run it again: all zeros. Two circuits, one gate apart, opposite statistics — the cheapest possible demonstration that the measurement you choose is half the physics.

My open questions

• What exactly does “collapse” mean physically — a process in the world, or an update to my description of it?— chasing
• How should I think about measurement when the state is mixed — and on entangled systems, where measuring one qubit changes what I can say about another?— next nodes
• What is the cleanest intuition for POVMs — what does a non-projective measurement look like physically?— open
• How does tomography actually reconstruct a state from measurement statistics — and how many measurements does it take?— bridges to benchmarking
• How does measurement error enter real hardware, and how is it separated from gate error?— bridges to the noise map

Explore further

• Nielsen & Chuang, Quantum Computation and Quantum Information, §2.2.3–2.2.6 — the measurement postulate, projective measurements, and POVMs in one sitting— rigor
• Matuschak & Nielsen, Quantum Computing for the Very Curious — the measurement sections build the “no peeking at amplitudes” intuition slowly and well— intuition
• Preskill, Ph219 lecture notes, ch. 3 — generalized measurement and POVMs, for when the projective case starts feeling too clean— deeper
• Qiskit — the |+⟩ experiment from Q4: measure, then H-then-measure, and compare the counts— hands on