Choosing a Measurement Basis

Measurement is always relative to a basis; the same state yields different statistics in different bases, and basis choice is a design lever.

Measurement needs a frame

There is no such thing as measuring a qubit in the abstract — every measurement is relative to a chosen orthonormal basis. The outcomes are the basis states, and their probabilities follow the Born rule. Choosing the basis is choosing which question you ask of the state.

Same state, different answers

Consider |+> = (|0>+|1>)/sqrt(2). Measured in the computational (Z) basis it gives 0 or 1 with equal probability — maximum randomness. Measured in the X basis it gives + every time — a definite result. The physical state is identical; only the question changed.

How to measure in another basis

Hardware natively measures in the computational basis. To measure in a different basis, apply a unitary that rotates the desired basis onto the computational one, then measure, then interpret. For the X basis, apply a Hadamard first; for the Y basis, apply S-dagger then H. The rotation and the measurement together implement the basis change.

• Z basis: measure directly
• X basis: apply H, then measure
• Y basis: apply S-dagger then H, then measure

Basis choice as a design lever

A well-designed algorithm ends in a basis where the answer is nearly deterministic, so few repetitions are needed. Poor basis choice can hide a computed result in 50/50 noise. Variational algorithms measure different observables by choosing corresponding bases, and error-correction reads parity checks by measuring in carefully chosen stabiliser bases.

The general rule

Any complete set of orthogonal states defines a valid measurement, and more general POVMs relax orthogonality to model realistic detectors. But for standard gate-based computing, the practical skill is knowing which single-qubit rotation turns the physically native Z measurement into the measurement your algorithm actually needs.