Quantum Measurement in Practice: From Amplitudes to Counts

A quantum program can finish with a state like this: equal parts zero and one. Run it once, though, and your screen shows only one result: 0 or 1. Run it a thousand times, and you might see 503 zeros and 497 ones. The gap between the state your program describes and the counts it returns is where quantum measurement becomes practical.

Amplitudes are not answers

For a single qubit, a pure state can be written as α|0⟩ + β|1⟩. The complex numbers α and β are amplitudes. They are not probabilities: the Born rule says the chance of observing 0 is |α|², and the chance of observing 1 is |β|². Those probabilities add to one for a normalized state.

For example, a Hadamard gate applied to |0⟩ creates (|0⟩ + |1⟩)/√2. Its two amplitudes are each 1/√2, so measuring in the computational basis gives each bit value a probability of one half. A simulator may show the state vector or exact probabilities, but a real measurement produces one sample, not a printed list of amplitudes.

With several qubits, each computational-basis bit string has an amplitude, and its probability is the squared magnitude of that amplitude. A program that measures a register therefore samples whole bit strings, such as 00, 01, 10, and 11. It does not independently flip a coin for every bit: correlations between qubits are part of the distribution.

Shots turn a distribution into a histogram

Quantum software commonly repeats the same circuit many times. Each repetition is called a shot. The results are collected into counts, which form a histogram: a practical estimate of the outcome probabilities.

In Qiskit, the measurement instruction connects a qubit to a classical bit so the result can be recorded. For example, the circuit operations below prepare a balanced qubit and request a computational-basis measurement:

qc.h(0)
qc.measure(0, 0)

The counts might look like {'0': 503, '1': 497} after 1,000 shots, but the exact split can change between runs. The circuit's ideal probabilities are still one half and one half. Counts are the observed sample, not a guarantee that the underlying probabilities have changed. When reading multi-qubit results, check how your framework orders displayed bit strings and maps classical bits to qubits.

Sampling has uncertainty

A finite histogram is noisy even when the circuit and device are perfect. For an outcome with probability p, the standard error of its observed fraction over N independent shots is approximately √(p(1-p)/N). For a balanced bit measured 1,000 times, that is about 0.016, or 1.6 percentage points. More shots generally shrink this sampling uncertainty in proportion to 1/√N; getting roughly half the error takes about four times as many shots.

This estimate describes random variation from sampling. Real devices add other effects, including gate errors, readout mistakes, and drift. More shots can make a noisy device's histogram more precise without making it closer to the ideal distribution. When comparing program results, separate these questions: how much might the counts vary just from finite sampling, and how faithfully does the hardware implement the intended circuit?

The measurement basis is a choice

“Computational basis” means the outcomes correspond to |0⟩ and |1⟩, often called the Z basis. Measuring in another basis changes which property is being sampled. A common trick is to rotate the state before a computational-basis measurement. For an X-basis measurement, apply a Hadamard gate immediately before measuring. The final readout is still 0 or 1, but the rotation makes those results encode the X-basis outcomes.

This matters because a state can look predictable in one basis and random in another. Before interpreting a histogram, ask what basis the circuit measures and what rotations happened just before readout. The measurement labels alone do not tell the whole story.

Collapse as an operational description

After measurement, it is useful to describe the qubit as having been projected onto the state matching the recorded result. If the result is 0, for instance, the post-measurement state is |0⟩ in the ideal projective model. This is often called collapse. In programming, the practical point is that measurement produces a classical record and changes the state available for later operations. It is not merely a command to inspect an unchanged hidden value.

A good debugging habit is to keep amplitudes, probabilities, and counts distinct. Amplitudes describe the state, the Born rule turns them into outcome probabilities for a chosen measurement, and repeated shots estimate those probabilities with finite uncertainty. Once these layers are clear, a histogram stops looking like an unpredictable program output and starts looking like what it is: evidence about a quantum distribution.