Quantum Mechanics for Computer Science

Chapter 5
Quantum Measurement and Statistical Interpretation

Diploma in Quantum Computing
Quantum Mechanics for Computer Science   •   Diploma

Quantum Measurement and Statistical Interpretation

Introduction

Quantum mechanics differs from classical physics most sharply in its account of measurement. A quantum state does not generally assign a definite pre-existing value to every observable. Instead, it determines probabilities for possible measurement outcomes. A measurement also changes the state, particularly when the measured observable is incompatible with a later observable.

For quantum computing, measurement is not an optional interpretation added after computation. It is the process by which information is extracted from qubits. Quantum algorithms therefore require a precise understanding of measurement probabilities, expectation values, post-measurement states, repeated trials, uncertainty, and disturbance.

This chapter develops the projective-measurement formalism used for finite-dimensional systems, especially qubits. The emphasis is on calculations and interpretation rather than philosophical debates about the meaning of the wavefunction.

Learning Objectives

By the end of this chapter, you should be able to:

  1. Explain the statistical interpretation of a quantum state.
  2. Calculate projective-measurement probabilities using projection operators.
  3. Determine expectation values and variances of observables.
  4. Calculate post-measurement states for selective and non-selective measurements.
  5. Interpret repeated measurements and empirical frequencies.
  6. Explain incompatible observables, uncertainty, and measurement disturbance.
  7. Apply measurement theory to a simple quantum-computing scenario.

1. States, Observables, and Outcomes

A pure state is represented by a normalized vector |ψ⟩ in a complex Hilbert space. Normalization means

⟨ψ|ψ⟩ = 1

A mixed state is represented by a density operator ρ satisfying

ρ = ρ†, ρ ≥ 0, Tr(ρ) = 1

Here ρ† is the conjugate transpose, ρ ≥ 0 means that all eigenvalues are non-negative, and Tr denotes the trace. For a pure state, ρ = |ψ⟩⟨ψ|.

An observable A is represented by a Hermitian operator:

A = A†

Hermiticity ensures that its eigenvalues are real, so they can represent physical measurement values. If A has distinct eigenvalues aₖ and corresponding orthogonal projectors Pₖ, then

A = ∑ₖ aₖ Pₖ

The projectors satisfy

PₖPⱼ = δₖⱼPₖ, ∑ₖ Pₖ = I

where δₖⱼ is 1 when k = j and 0 otherwise, and I is the identity operator. The equations assume a finite-dimensional system and a discrete projective measurement.

For a non-degenerate eigenstate |aₖ⟩, the projector is Pₖ = |aₖ⟩⟨aₖ|. If an eigenvalue is degenerate, Pₖ projects onto the entire eigenspace associated with that eigenvalue.

2. The Born Rule

The Born rule gives the probability of obtaining outcome aₖ when observable A is measured in state ρ:

p(aₖ) = Tr(Pₖρ)

For a pure state |ψ⟩, this becomes

p(aₖ) = ⟨ψ|Pₖ|ψ⟩ = ||Pₖ|ψ⟩||²

The probability is dimensionless and lies between 0 and 1. Completeness of the projectors guarantees that probabilities sum to one:

∑ₖ p(aₖ) = ∑ₖ Tr(Pₖρ) = Tr(ρ) = 1

For a qubit measured in the computational basis {|0⟩, |1⟩}, the projectors are

P₀ = |0⟩⟨0|, P₁ = |1⟩⟨1|

If

|ψ⟩ = α|0⟩ + β|1⟩

with |α|² + |β|² = 1, then

p(0) = |α|², p(1) = |β|²

The complex phases of α and β affect interference in later operations, but the computational-basis probabilities depend on their squared magnitudes.

Worked Example 1: Computational-Basis Measurement

Consider

|ψ⟩ = (√3/2)|0⟩ + (1/2)|1⟩

The state is normalized because 3/4 + 1/4 = 1. A computational-basis measurement gives

p(0) = |√3/2|² = 3/4

and

p(1) = |1/2|² = 1/4

Thus, in a large number of identical trials, approximately 75% of outcomes are 0 and 25% are 1.

3. State Update After Measurement

The probability rule describes the outcome distribution, but a complete measurement model must also describe the state after an outcome occurs.

For a projective measurement with outcome associated with Pₖ, the selective post-measurement state is

ρₖ = PₖρPₖ / Tr(Pₖρ)

This formula applies when p(aₖ) = Tr(Pₖρ) is non-zero. For a pure state, the equivalent expression is

|ψₖ⟩ = Pₖ|ψ⟩ / √⟨ψ|Pₖ|ψ⟩

The denominator normalizes the projected state. In the non-degenerate case, an ideal projective measurement leaves the system in the corresponding eigenstate |aₖ⟩.

A selective measurement records the particular outcome. If the result is not recorded, the appropriate state is the non-selective post-measurement state:

ρ′ = ∑ₖ PₖρPₖ

This state is generally mixed, even if ρ was initially pure. It contains the statistical state after the measurement apparatus has produced an outcome but the observer has discarded the outcome record.

Worked Example 2: Post-Measurement State

Let

|ψ⟩ = (|0⟩ + |1⟩)/√2

A computational-basis measurement has probabilities p(0) = p(1) = 1/2. If outcome 0 is obtained, the unnormalized projected vector is

P₀|ψ⟩ = |0⟩/√2

After normalization, the state is |0⟩. Similarly, outcome 1 leaves the state |1⟩.

If the result is ignored, the density operator becomes

ρ′ = (1/2)|0⟩⟨0| + (1/2)|1⟩⟨1|

The off-diagonal coherence terms have disappeared. This is called dephasing in the computational basis.

4. Expectation Values and Variance

The expectation value is the long-run average of an observable over many identically prepared systems. For observable A and state ρ,

⟨A⟩ = Tr(ρA)

The equivalent pure-state expression is

⟨A⟩ = ⟨ψ|A|ψ⟩

Using the spectral decomposition A = ∑ₖ aₖPₖ gives

⟨A⟩ = ∑ₖ aₖp(aₖ)

The expectation value need not itself be a possible single measurement outcome. For example, a spin observable may produce only +1 or −1 in suitable units, while its expectation value may be 0.3.

The variance measures the spread of outcomes:

(∆A)² = ⟨A²⟩ − ⟨A⟩²

The standard deviation is

∆A = √(⟨A²⟩ − ⟨A⟩²)

The variance is non-negative because it is the average squared deviation from the mean. These formulas assume a normalized state and a Hermitian observable.

Worked Example 3: Pauli-Z Expectation Value

The Pauli-Z operator is

Z = |0⟩⟨0| − |1⟩⟨1| = [1 0; 0 −1]

For the state

|ψ⟩ = √3/2|0⟩ + 1/2|1⟩

the probabilities of the Z outcomes +1 and −1 are 3/4 and 1/4. Therefore,

⟨Z⟩ = (+1)(3/4) + (−1)(1/4) = 1/2

Since Z² = I,

⟨Z²⟩ = 1

Thus,

(∆Z)² = 1 − (1/2)² = 3/4

and

∆Z = √3/2

5. Repeated Trials and Statistical Interpretation

A single quantum measurement produces one outcome, not an expectation value. Probabilities become empirically meaningful through repeated preparation and measurement of identically prepared systems.

If an outcome has probability p and the experiment is repeated N independent times, the number n of occurrences is described by a binomial distribution:

Pr(n) = C(N,n)pⁿ(1−p)ᴺ⁻ⁿ

where C(N,n) = N!/[n!(N−n)!]. The sample frequency n/N approaches p as N increases. Its standard deviation is

σfrequency = √[p(1−p)/N]

This expression assumes independent trials with the same probability p. It shows that statistical uncertainty decreases approximately as 1/√N, not as 1/N.

For p = 1/2 and N = 10,000, the standard deviation of the frequency is

√[(1/2)(1/2)/10,000] = 0.005

or 0.5 percentage points. A measured frequency near 50% is therefore expected, but exact equality is not required.

Quantum randomness is not normally interpreted as ignorance about a definite classical outcome. The theory specifies probabilities, and individual outcomes are irreducibly probabilistic under the standard interpretation.

6. Qubit Measurements in Different Bases

The computational basis is not the only measurement basis. The Pauli-X operator has eigenstates

|+⟩ = (|0⟩ + |1⟩)/√2, |−⟩ = (|0⟩ − |1⟩)/√2

with eigenvalues +1 and −1. Measuring in the X basis uses projectors

P₊ = |+⟩⟨+|, P₋ = |−⟩⟨−|

For a general state |ψ⟩ = α|0⟩ + β|1⟩,

p(+) = |⟨+|ψ⟩|² = |(α + β)/√2|²

and

p(−) = |⟨−|ψ⟩|² = |(α − β)/√2|²

The relative phase between α and β affects these probabilities. Consequently, two states can have the same computational-basis probabilities but different X-basis probabilities.

In laboratory and quantum-computing practice, an X-basis measurement can be implemented by applying a Hadamard gate H and then measuring in the computational basis. This works because H maps |+⟩ to |0⟩ and |−⟩ to |1⟩.

7. Compatible and Incompatible Observables

Two observables A and B are compatible when they commute:

[A,B] = AB − BA = 0

For finite-dimensional projective measurements, commuting observables can be simultaneously diagonalized, allowing joint definite values in a common eigenbasis.

The Pauli operators X and Z do not commute:

[X,Z] = −2iY

where Y is the Pauli-Y operator. Therefore, X and Z are incompatible. An X measurement generally changes the probabilities of a subsequent Z measurement, and vice versa.

For a qubit in |0⟩, a Z measurement is certain to produce +1 and leaves the state unchanged. However, an X measurement produces +1 or −1 with equal probability and leaves the system in |+⟩ or |−⟩. A later Z measurement then produces either result with equal probability. This is measurement disturbance, not merely additional classical uncertainty.

8. Uncertainty Relations

For observables A and B, the Robertson uncertainty relation is

∆A∆B ≥ (1/2)|⟨[A,B]⟩|

The expectation value in the right-hand side is taken in the same state as the uncertainties. The relation assumes A and B are Hermitian observables and the state has finite variances.

A stronger Schrödinger form is

(∆A)²(∆B)² ≥ (1/4)|⟨[A,B]⟩|² + (1/4)|⟨{∆A,∆B}⟩|²

where ∆A = A − ⟨A⟩I, ∆B = B − ⟨B⟩I, and {R,S} = RS + SR is the anticommutator. The simpler Robertson relation is usually sufficient for introductory qubit analysis.

For X and Z, [X,Z] = −2iY. Hence

∆X∆Z ≥ |⟨Y⟩|

in units where the Pauli eigenvalues are ±1. For the state |+⟩, ∆X = 0 and ⟨Y⟩ = 0, so the bound is 0; this is consistent because |+⟩ is an eigenstate of X. The state has definite X but completely uncertain Z.

Uncertainty is not simply a claim that the apparatus is inaccurate. It is a structural feature of noncommuting quantum observables.

9. Measurement Disturbance and Quantum Information

A measurement can remove phase information even when the outcome is not used. Consider a general qubit density matrix:

ρ = [ρ₀₀ ρ₀₁; ρ₁₀ ρ₁₁]

A computational-basis measurement without retaining the result produces

ρ′ = [ρ₀₀ 0; 0 ρ₁₁]

The diagonal populations remain, while the off-diagonal coherence terms vanish. This is why measurement is generally irreversible: the original phase relation cannot be reconstructed from the non-selective state alone.

In quantum algorithms, measurement is normally postponed until the end because intermediate measurements may destroy interference needed for the computation. Nevertheless, measurement can be useful in error correction, quantum teleportation, state preparation, and adaptive circuits. In these applications, classical measurement outcomes may control later quantum operations.

The measurement basis must therefore be selected according to the information required. Measuring a state in the wrong basis can destroy precisely the correlations or phase information that an algorithm needs.

10. Illustrative Case: Estimating a Qubit State

Background

A research team is testing a quantum processor that prepares an unknown single-qubit state. They wish to estimate its behavior in the three Pauli measurement bases X, Y, and Z.

Problem

The team cannot determine the complete state from one measurement. They need repeated trials and must avoid assuming that computational-basis data reveal all quantum information.

Approach

They prepare the same state many times. To measure Z, they measure directly in the computational basis. To measure X, they apply H and then measure computationally. To measure Y, they apply a basis-change operation appropriate to the Y eigenstates and then measure computationally.

For each basis, they estimate the expectation value using

⟨A⟩estimated = (N₊ − N₋)/(N₊ + N₋)

where N₊ and N₋ are the counts of the +1 and −1 outcomes. This formula applies to a two-outcome observable with eigenvalues ±1.

Suppose the Z experiment gives 7,500 positive outcomes and 2,500 negative outcomes out of 10,000 trials. Then

⟨Z⟩estimated = (7,500 − 2,500)/10,000 = 0.5

The estimated probability of the positive Z outcome is 0.75. Similar estimates for X and Y determine the Bloch-vector components, provided experimental errors are controlled.

Outcome

The team obtains an operational description of the state from repeated data rather than from a single shot. Their results can be used to compare the prepared state with the intended state and to diagnose noise, calibration errors, or unwanted dephasing.

Learning Points

  • A single shot cannot estimate a probability reliably.
  • Different measurement bases reveal different components of a quantum state.
  • Measurement statistics are essential for validating quantum hardware.
  • Finite sample size creates statistical uncertainty even when the device is ideal.

11. Limitations and Practical Considerations

The projective-measurement model is fundamental but not universal. Real detectors may be inefficient, noisy, or biased. More general measurements are described by positive-operator-valued measures (POVMs), whose elements Eₖ satisfy

Eₖ ≥ 0, ∑ₖ Eₖ = I

The probability rule becomes

p(k) = Tr(Eₖρ)

A POVM element need not be a projector. This formalism models noisy detection, unsharp measurements, and many quantum-information protocols. The state-update operation for a POVM requires measurement operators Mₖ satisfying Eₖ = Mₖ†Mₖ; the post-measurement state is MₖρMₖ†/p(k). Different physical instruments can produce the same probabilities but different state disturbances.

In practical quantum computing, measurement results are also affected by readout error. If the actual outcome is 0 but the detector reports 1 with probability ε₀₁, and the actual outcome is 1 but the detector reports 0 with probability ε₁₀, observed frequencies must be corrected using a calibrated readout model. Such correction cannot restore information that was never detected reliably, and it can amplify statistical noise.

Best Practices for Measurement Analysis

  1. Normalize the state before calculating probabilities.
  2. Identify the measurement basis and construct its projectors.
  3. Use p(k) = Tr(Pₖρ) for density operators and p(k) = ⟨ψ|Pₖ|ψ⟩ for pure states.
  4. Check that all probabilities are non-negative and sum to one.
  5. Normalize the projected state after a selective outcome.
  6. Distinguish a single outcome from an expectation value.
  7. Use sufficiently many repeated trials to estimate probabilities.
  8. Check commutators when comparing sequential measurements.
  9. Account for detector noise and finite-sample uncertainty in experiments.
  10. Do not infer phase information from one measurement basis alone.

Chapter Summary

Quantum measurement connects the mathematical state of a system with experimentally observed outcomes. An observable is represented by a Hermitian operator, and its projectors determine the possible outcomes. The Born rule gives p(aₖ) = Tr(Pₖρ), while a selective projective measurement updates the state to PₖρPₖ/Tr(Pₖρ). If the outcome is ignored, the state becomes ∑ₖPₖρPₖ.

Expectation values are long-run averages, ⟨A⟩ = Tr(ρA), and variances quantify outcome spread. Repeated trials reveal probabilities through frequencies, with statistical uncertainty decreasing approximately as 1/√N. Measurements in different bases reveal different aspects of a state, including relative phase information.

Commuting observables are compatible; noncommuting observables are incompatible and may disturb one another. The Robertson uncertainty relation quantifies a fundamental limit on simultaneous sharpness. In quantum computing, measurement basis selection, state disturbance, readout error, and statistical sampling are central to algorithm design and hardware validation. Projective measurements provide the essential foundation, while POVMs extend the theory to realistic and generalized measurement devices.

Key Terms

Born rule: Rule assigning probabilities to quantum measurement outcomes.

Observable: Hermitian operator representing a measurable physical quantity.

Projector: Operator that extracts the component of a state in a specified subspace.

Projective measurement: Measurement described by mutually orthogonal projectors summing to the identity.

Density operator: Operator representation of a pure or mixed quantum state.

Pure state: State representable by one normalized state vector.

Mixed state: Statistical mixture of possible quantum states.

Expectation value: Probability-weighted average of measurement outcomes.

Variance: Mean squared deviation of outcomes from the expectation value.

Standard deviation: Square root of the variance.

Selective measurement: Measurement in which the observed outcome is retained.

Non-selective measurement: Measurement in which the outcome record is ignored.

Measurement disturbance: Change in a quantum state caused by measurement.

Compatible observables: Observables whose operators commute.

Incompatible observables: Observables whose operators do not commute.

Uncertainty relation: Bound on the joint sharpness of two observables.

Eigenstate: State that an observable maps to a scalar multiple of itself.

POVM: Generalized measurement described by positive operators summing to identity.

Dephasing: Loss of off-diagonal coherence in a chosen basis.

Readout error: Difference between the actual quantum outcome and the detector report.

Review Questions

Short Answer Questions

  1. State the Born rule for a projective measurement on a density operator.
  2. What is the difference between a selective and a non-selective measurement?
  3. Define the expectation value of an observable.
  4. What condition makes two observables compatible?
  5. Why are repeated trials required to estimate a quantum probability experimentally?

Long Answer Questions

  1. Explain projective measurement using projectors, including the probability rule and state-update rule.
  2. Derive the computational-basis probabilities for a general qubit α|0⟩ + β|1⟩.
  3. Explain expectation value, variance, and standard deviation, and show how they relate to measurement statistics.
  4. Discuss how measurement in the X basis differs from measurement in the computational basis.
  5. Explain the uncertainty relation and measurement disturbance using the Pauli X and Z observables.

Analytical / Application Questions

  1. A qubit is prepared in |ψ⟩ = (√3/2)|0⟩ + (1/2)|1⟩. Calculate the probabilities, expectation value, and standard deviation for a computational-basis measurement interpreted as a Z measurement.
  2. A state is measured in the computational basis, but the outcome is discarded. Explain the resulting density operator and identify which information has been lost.
  3. Design a repeated-trial procedure for estimating the expectation value of a two-outcome observable with eigenvalues +1 and −1. State how sample size and detector errors affect the estimate.

Multiple Choice Questions with Answers

Question 1

For a projective measurement with projector Pₖ on state ρ, what is the probability of outcome k?

A. Tr(Pₖρ)

B. Tr(Pₖ + ρ)

C. Tr(ρ)/Tr(Pₖ)

D. det(Pₖρ)

Correct Answer: A. Tr(Pₖρ)

Question 2

What is the computational-basis probability of obtaining 1 from |ψ⟩ = α|0⟩ + β|1⟩?

A. |α|

B. |β|²

C. αβ

D. |α + β|²

Correct Answer: B. |β|²

Question 3

Which expression gives the expectation value of observable A in state ρ?

A. Tr(ρA)

B. Tr(ρ + A)

C. det(ρA)

D. Tr(A) − Tr(ρ)

Correct Answer: A. Tr(ρA)

Question 4

If a selective projective measurement produces outcome k, what is the updated state?

A. PₖρPₖ / Tr(Pₖρ)

B. ρ + Pₖ

C. Pₖρ / Tr(ρ)

D. ρPₖρ

Correct Answer: A. PₖρPₖ / Tr(Pₖρ)

Question 5

For N independent trials with outcome probability p, how does the standard deviation of the observed frequency scale with N?

A. N

B. 1/N²

C. 1/√N

D. √N

Correct Answer: C. 1/√N

Question 6

Two observables are compatible in the projective-measurement framework when:

A. They have identical eigenvalues

B. Their commutator is zero

C. Their matrices have the same trace

D. They are both measured once

Correct Answer: B. Their commutator is zero

Question 7

What happens when the state |+⟩ is measured in the computational basis?

A. Outcome 0 occurs with certainty

B. Outcome 1 occurs with certainty

C. Both outcomes occur with probability 1/2

D. No outcome is possible

Correct Answer: C. Both outcomes occur with probability 1/2

Question 8

For the Pauli-Z observable, whose eigenvalues are +1 and −1, what is Z²?

A. 0

B. Z

C. −I

D. I

Correct Answer: D. I

Question 9

What is the non-selective post-measurement state for projectors {Pₖ}?

A. ∑ₖPₖρPₖ

B. ∏ₖPₖρ

C. ρ²

D. ∑ₖPₖ + ρ

Correct Answer: A. ∑ₖPₖρPₖ

Question 10

A measurement in a basis incompatible with a later measurement can change the later outcome probabilities because it:

A. Always increases the energy

B. Disturbs the quantum state

C. Changes the dimension of the Hilbert space

D. Makes all outcomes deterministic

Correct Answer: B. Disturbs the quantum state
Quantum Mechanics for Computer Science
Chapter 5 of 14