39 min read

Classical physics permits an observer to measure a system without disturbing it. You can weigh a rock, clock a sprinter, or read a voltmeter, and the act of measurement leaves the system essentially unchanged. Quantum mechanics offers no such...

In This Chapter

Exercises Quiz Case Study 01 Case Study 02 Key Takeaways Further Reading

Chapter 4: Measurement — The Born Rule, Projection, Collapse, and Why Observing a Qubit Changes It

Learning Objectives

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

  • State the Born rule and compute measurement outcome probabilities for arbitrary pure states.
  • Describe projective (von Neumann) measurements in terms of Hermitian observables and their spectral decompositions.
  • Explain the collapse postulate and its consequences for subsequent measurements.
  • Determine whether two quantum states are distinguishable by a single measurement.
  • Define POVMs and explain why they generalize projective measurements.
  • Prove the no-cloning theorem and articulate its implications for quantum information.
  • Outline the procedure of quantum state tomography.
  • Use Qiskit to simulate measurement statistics and verify theoretical predictions.
  • Derive the Born rule from consistency arguments and connect it to the geometry of Hilbert space.
  • Compute expectation values, variances, and uncertainty relations for quantum observables.
  • Construct POVMs for optimal unambiguous state discrimination and calculate their success probabilities.
  • Implement full quantum state tomography and interpret the results in terms of fidelity.

4.1 The Measurement Problem

Classical physics permits an observer to measure a system without disturbing it. You can weigh a rock, clock a sprinter, or read a voltmeter, and the act of measurement leaves the system essentially unchanged. Quantum mechanics offers no such luxury. Measurement is not a passive readout; it is an active intervention that irreversibly alters the state.

This chapter confronts the measurement problem head-on. We will develop the mathematical machinery that connects the abstract state vector $|\psi\rangle$ to the concrete numbers that appear on laboratory instruments. Along the way, we will encounter some of the most philosophically charged ideas in all of science — collapse, indistinguishability, and the impossibility of copying quantum information.

Recurring Theme — The Observer Effect: In quantum computing, measurement is not a bug to be avoided but a resource to be harnessed. Every quantum algorithm ends with measurement. Error correction relies on syndrome measurements. Quantum cryptography exploits the fact that measurement disturbs. Mastering measurement is mastering the interface between quantum and classical worlds.

Historical Context: From Planck to Born

The measurement problem has roots in the earliest days of quantum mechanics. In 1900, Max Planck introduced the quantum hypothesis to explain blackbody radiation, but it was not until 1926 that Max Born provided the probabilistic interpretation of the wavefunction that bears his name. Born's insight — that the square of the amplitude gives the probability — transformed Schrödinger's wave equation from a deterministic evolution of a probability wave into a tool for predicting measurement outcomes. This was not a minor technical detail; it was a conceptual revolution. Classical mechanics predicts what will happen. Quantum mechanics predicts how likely things are to happen.

Born later recalled that he was influenced by Einstein's idea of a "ghost field" (Gespensterfeld), where the field amplitude determines the probability of a particle's detection. Born extended this: the wavefunction $\psi$ itself is not a physical field, but its modulus squared $|\psi|^2$ gives the probability density. For this insight, Born received the Nobel Prize in Physics in 1954 — twenty-eight years after his original paper.

Common Misconception: "Measurement collapses the wavefunction because we interact with it physically." This is a subtle error. The collapse postulate is not a consequence of physical interaction in the ordinary sense. A strong laser pulse interacting with an atom can be modeled unitarily — it is the extraction of classical information that triggers collapse in the standard interpretation. The boundary between unitary evolution and collapse is precisely what the measurement problem is about. In quantum computing, we sidestep the philosophical debate: measurement is an operational primitive that yields a classical outcome and updates our state of knowledge.

4.2 The Born Rule

4.2.1 Statement

The Born rule, formulated by Max Born in 1926, is the fundamental link between the mathematical formalism of quantum mechanics and experimental outcomes. It is a probabilistic rule — quantum mechanics predicts probabilities, not certainties.

Born Rule (Projective Form). Let a quantum system be in the pure state $|\psi\rangle$, and let $\{|e_k\rangle\}$ be an orthonormal basis of the Hilbert space. The probability of obtaining the outcome corresponding to basis state $|e_k\rangle$ when measuring in this basis is:

$$P(k) = |\langle e_k | \psi \rangle|^2$$

More generally, if we measure an observable $A$ with spectral decomposition:

$$A = \sum_k a_k |a_k\rangle\langle a_k|$$

where $a_k$ are the eigenvalues and $|a_k\rangle$ the corresponding eigenstates, then the probability of obtaining outcome $a_k$ is:

$$P(a_k) = |\langle a_k | \psi \rangle|^2 = \langle \psi | a_k \rangle \langle a_k | \psi \rangle = \langle \psi | P_k | \psi \rangle$$

where $P_k = |a_k\rangle\langle a_k|$ is the projector onto the eigenspace of $a_k$.

4.2.2 Derivation from Consistency: Why Squared Amplitudes?

One might ask: why squared amplitudes rather than, say, amplitudes themselves or fourth powers? The answer lies in the structure of probability theory and the geometry of Hilbert space.

Consider a state $|\psi\rangle = \alpha|0\rangle + \beta|1\rangle$ with $|\alpha|^2 + |\beta|^2 = 1$. We require that:

  1. Probabilities are non-negative and sum to one. Any function $f$ mapping amplitudes to probabilities must satisfy $f(|\alpha|^2) + f(|\beta|^2) = 1$ for all valid states. The simplest choice is $f(x) = x$, giving $P(k) = |\langle e_k|\psi\rangle|^2$.

  2. Composition rule for sequential measurements. If we measure in basis $\{|e_k\rangle\}$ and then in basis $\{|f_j\rangle\}$, the probability of getting $k$ then $j$ must equal $P(k) \cdot P(j|k)$. This chain rule is only consistent with $P(k) = |\langle e_k|\psi\rangle|^2$.

  3. Gleason's theorem (1957). In dimensions 3 and above, the only probability measure on the lattice of projection operators that is invariant under unitary transformations is $P(P_k) = \mathrm{Tr}(P_k \rho)$, which reduces to $|\langle e_k|\psi\rangle|^2$ for pure states. Gleason's theorem is a deep result: it says the Born rule is the only consistent assignment of probabilities in Hilbert spaces of dimension $\geq 3$.

The squared-modulus rule also has a beautiful geometric interpretation. The amplitude $\langle e_k|\psi\rangle$ is the component of $|\psi\rangle$ along $|e_k\rangle$ — essentially, the "shadow" of $|\psi\rangle$ on $|e_k\rangle$. The probability is the squared length of this projection, which by Pythagoras's theorem generalizes to:

$$\sum_k |\langle e_k|\psi\rangle|^2 = \langle\psi|\psi\rangle = 1$$

The normalization condition ensures that probabilities sum to one — a reassuring consistency check that follows directly from the completeness of the basis.

4.2.3 Example: Measuring a Qubit in the Computational Basis

Consider a single qubit in the state:

$$|\psi\rangle = \alpha|0\rangle + \beta|1\rangle, \quad |\alpha|^2 + |\beta|^2 = 1$$

Measuring in the computational basis $\{|0\rangle, |1\rangle\}$ yields:

$$P(0) = |\langle 0|\psi\rangle|^2 = |\alpha|^2$$ $$P(1) = |\langle 1|\psi\rangle|^2 = |\beta|^2$$

Worked Example 4.1: Let $|\psi\rangle = \frac{1}{\sqrt{3}}|0\rangle + \sqrt{\frac{2}{3}}i|1\rangle$. Then:

$$P(0) = \left|\frac{1}{\sqrt{3}}\right|^2 = \frac{1}{3}$$ $$P(1) = \left|\sqrt{\frac{2}{3}}i\right|^2 = \frac{2}{3}|i|^2 = \frac{2}{3}$$

Check: $\frac{1}{3} + \frac{2}{3} = 1$. The global phase of the amplitude is irrelevant for probabilities, but the relative phase between amplitudes matters when interference is involved.

Worked Example 4.2: Let $|\psi\rangle = \frac{3}{5}|0\rangle + \frac{4}{5}e^{i\pi/3}|1\rangle$. Then:

$$P(0) = \left(\frac{3}{5}\right)^2 = \frac{9}{25}$$ $$P(1) = \left(\frac{4}{5}\right)^2 = \frac{16}{25}$$

The phase $e^{i\pi/3}$ has no effect on the probabilities in the computational basis — it only matters if we later measure in a different basis or apply further unitary operations.

Worked Example 4.3: Consider a state on the equator of the Bloch sphere: $|\psi\rangle = \frac{1}{\sqrt{2}}(|0\rangle + e^{i\phi}|1\rangle)$. Regardless of the value of $\phi$:

$$P(0) = P(1) = \frac{1}{2}$$

All equatorial states give 50-50 outcomes in the Z-basis. The phase $\phi$ only reveals itself when we measure in the X or Y bases, or when the state interferes with itself in a quantum circuit.

Try It Yourself: Compute $P(0)$ and $P(1)$ for the state $|\psi\rangle = \frac{2}{3}|0\rangle + \frac{\sqrt{5}}{3}e^{-i\pi/6}|1\rangle$. Verify that probabilities sum to 1.

4.2.4 Example: Measuring in the Hadamard Basis

The same qubit measured in the Hadamard basis $\{|+\rangle, |-\rangle\}$ where:

$$|+\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle), \quad |-\rangle = \frac{1}{\sqrt{2}}(|0\rangle - |1\rangle)$$

yields:

$$P(+) = |\langle +|\psi\rangle|^2 = \frac{1}{2}|\alpha + \beta|^2$$ $$P(-) = |\langle -|\psi\rangle|^2 = \frac{1}{2}|\alpha - \beta|^2$$

Notice that the same physical state produces different probability distributions depending on which basis we choose to measure. This is a deep point: measurement outcomes are not properties of the system alone but of the system and the measurement apparatus together.

Worked Example 4.4: Let $|\psi\rangle = \frac{1}{\sqrt{3}}|0\rangle + \sqrt{\frac{2}{3}}|1\rangle$. Then:

$$P(+) = \frac{1}{2}\left|\frac{1}{\sqrt{3}} + \sqrt{\frac{2}{3}}\right|^2 = \frac{1}{2}\left(\frac{1}{\sqrt{3}} + \sqrt{\frac{2}{3}}\right)^2 = \frac{1}{2}\left(\frac{1}{3} + \frac{2}{3} + \frac{2\sqrt{2}}{3}\right) = \frac{1}{2} + \frac{\sqrt{2}}{3}$$

$$P(-) = \frac{1}{2}\left|\frac{1}{\sqrt{3}} - \sqrt{\frac{2}{3}}\right|^2 = \frac{1}{2} - \frac{\sqrt{2}}{3}$$

Check: $\frac{1}{2} + \frac{\sqrt{2}}{3} + \frac{1}{2} - \frac{\sqrt{2}}{3} = 1$. When $\alpha$ and $\beta$ have the same sign, the $|+\rangle$ outcome is more likely — constructive interference in the Hadamard basis.

Worked Example 4.5: For $|\psi\rangle = |+\rangle$ itself, measuring in the Hadamard basis:

$$P(+) = |\langle +|+\rangle|^2 = 1, \quad P(-) = |\langle -|+\rangle|^2 = 0$$

The state is an eigenstate of $X$ (the Hadamard-basis observable), so the outcome is deterministic. This illustrates a general principle: measuring in the eigenbasis of an observable always yields a deterministic outcome for eigenstates.

# Qiskit demonstration: Born rule probabilities for a superposition state
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
import numpy as np

# Create a qubit in state (1/sqrt(2))(|0> + |1>)
qc = QuantumCircuit(1, 1)
qc.h(0)          # Hadamard creates equal superposition
qc.measure(0, 0)

# Simulate 10,000 shots
simulator = AerSimulator()
job = simulator.run(qc, shots=10000)
counts = job.result().get_counts()

print("Measurement counts:", counts)
print("P(|0>) ≈", counts.get('0', 0) / 10000)
print("P(|1>) ≈", counts.get('1', 0) / 10000)
print("Theoretical: P(0) = 0.5, P(1) = 0.5")

# Qiskit: Born rule verification with arbitrary amplitudes
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
from qiskit.quantum_info import Statevector
import numpy as np

# Prepare state (1/sqrt(3))|0> + sqrt(2/3)*i|1>
alpha = 1/np.sqrt(3)
beta = np.sqrt(2/3) * 1j

# Use U3 gate to prepare the state
# |psi> = cos(theta/2)|0> + e^{i*phi} sin(theta/2)|1>
# cos(theta/2) = 1/sqrt(3), phi = pi/2, lambda = 0
theta = 2 * np.arccos(np.abs(alpha))
phi = np.angle(beta)
lam = 0

qc = QuantumCircuit(1, 1)
qc.u(theta, phi, lam, 0)
qc.measure(0, 0)

simulator = AerSimulator()
job = simulator.run(qc, shots=10000)
counts = job.result().get_counts()
p0 = counts.get('0', 0) / 10000
p1 = counts.get('1', 0) / 10000

print(f"Empirical P(|0>) = {p0:.4f}, theoretical = {1/3:.4f}")
print(f"Empirical P(|1>) = {p1:.4f}, theoretical = {2/3:.4f}")

4.2.5 Measuring in an Arbitrary Basis: The General Procedure

Given an arbitrary orthonormal basis $\{|v_0\rangle, |v_1\rangle\}$ of $\mathbb{C}^2$, how do we implement a measurement in that basis on a quantum computer? The answer is elegant: rotate the basis into the computational basis, then measure.

If $\{|v_0\rangle, |v_1\rangle\}$ is related to $\{|0\rangle, |1\rangle\}$ by a unitary $U$ (i.e., $|v_k\rangle = U|k\rangle$), then measuring in the $\{|v_k\rangle\}$ basis is equivalent to:

  1. Apply $U^\dagger$ to the state
  2. Measure in the computational basis
  3. Relabel outcome $k$ as corresponding to $|v_k\rangle$

This follows because $|\langle v_k|\psi\rangle|^2 = |\langle k|U^\dagger|\psi\rangle|^2$.

Example: Y-basis measurement. The Y-basis is $\{|+i\rangle, |-i\rangle\}$ where $|+i\rangle = \frac{1}{\sqrt{2}}(|0\rangle + i|1\rangle)$ and $|-i\rangle = \frac{1}{\sqrt{2}}(|0\rangle - i|1\rangle)$. The unitary mapping $\{|0\rangle, |1\rangle\} \to \{|+i\rangle, |-i\rangle\}$ is $S^\dagger H$ (Hadamard followed by $S^\dagger$ in reverse). So to measure in the Y-basis:

  1. Apply $H S^\dagger$ (or equivalently $S H$ in the proper order: first $S^\dagger$, then $H$)
  2. Measure in the computational basis
ASCII Art: Measuring in Different Bases

Z-basis:  ──[M]──             (direct measurement)
X-basis:  ──[H]──[M]──        (rotate to Z, then measure)
Y-basis:  ──[S†]──[H]──[M]──  (rotate to Z, then measure)

General:  ──[U†]──[M]──        (rotate basis to computational, measure)

Common Misconception: "The Born rule only applies to orthogonal measurements." Strictly speaking, the Born rule as stated above applies to projective measurements, which are defined by orthogonal projectors. However, the generalized Born rule for POVMs (Section 4.5) extends probability calculations to non-orthogonal measurements. The core principle remains: probability equals the squared overlap between state and measurement element.

4.2.6 The Geometry of the Born Rule: Projection as Shadow

The Born rule has a beautiful geometric interpretation. If we think of the state $|\psi\rangle$ as a vector in Hilbert space and the basis state $|e_k\rangle$ as a direction, then $P(k) = |\langle e_k|\psi\rangle|^2$ is the squared length of the "shadow" that $|\psi\rangle$ casts onto $|e_k\rangle$.

ASCII Art: Born Rule as Projection

       |ψ⟩
       /|
      / |
     /  |  ⟨e_k|ψ⟩ = projection of |ψ⟩ onto |e_k⟩
    /   |
   /    |
  /_____|
  |e_k⟩

  P(k) = |⟨e_k|ψ⟩|² = (length of shadow)²
  Σ_k P(k) = 1  (Pythagoras's theorem in Hilbert space)

This geometric picture makes several facts intuitive:

  1. Probabilities sum to one: $\sum_k |\langle e_k|\psi\rangle|^2 = \langle\psi|\psi\rangle = 1$ by Parseval's theorem (Pythagoras in higher dimensions).

  2. Orthogonal bases give different probabilities: Rotating the "axes" (changing the measurement basis) redistributes the "shadow lengths" while preserving their squared sum.

  3. Eigenstates give deterministic outcomes: If $|\psi\rangle = |e_k\rangle$ (pointing along one axis), the entire shadow falls on that axis: $P(k) = 1$, all others zero.

Worked Example 4.10: Consider $|\psi\rangle = \cos(\pi/6)|0\rangle + \sin(\pi/6)|1\rangle = \frac{\sqrt{3}}{2}|0\rangle + \frac{1}{2}|1\rangle$. In the computational basis:

$$P(0) = \cos^2(\pi/6) = \frac{3}{4}, \quad P(1) = \sin^2(\pi/6) = \frac{1}{4}$$

Now rotate to the $X$-basis. The projections are:

$$\langle +|\psi\rangle = \frac{1}{\sqrt{2}}(\cos(\pi/6) + \sin(\pi/6)) = \frac{1}{\sqrt{2}} \cdot \frac{\sqrt{3}+1}{2}$$

$$P(+) = \frac{1}{2}\left(\frac{\sqrt{3}+1}{2}\right)^2 = \frac{4 + 2\sqrt{3}}{8} = \frac{2 + \sqrt{3}}{4} \approx 0.933$$

The "shadow" has shifted — most of the probability mass is now on the $|+\rangle$ outcome, because $|\psi\rangle$ is closer to $|+\rangle$ than to $|-\rangle$ on the Bloch sphere.

Try It Yourself: For $|\psi\rangle = \frac{1}{2}|0\rangle + \frac{\sqrt{3}}{2}|1\rangle$, compute $P(+)$, $P(-)$, and verify they sum to 1. How does the result compare to the example above? What does this tell you about the relationship between the Bloch sphere angle and the measurement probabilities?

4.2.7 The Born Rule for Mixed States

The Born rule extends naturally to mixed states described by density matrices $\rho$. If $\rho$ is a density operator and $\{|e_k\rangle\}$ is an orthonormal basis, then:

$$P(k) = \langle e_k|\rho|e_k\rangle = \text{Tr}(\rho |e_k\rangle\langle e_k|) = \text{Tr}(\rho P_k)$$

This is the most general form of the Born rule. For pure states $\rho = |\psi\rangle\langle\psi|$, it reduces to $P(k) = |\langle e_k|\psi\rangle|^2$.

Worked Example 4.11: The maximally mixed state $\rho = \frac{1}{2}I = \frac{1}{2}\begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$.

$$P(0) = \text{Tr}(\rho |0\rangle\langle 0|) = \langle 0|\rho|0\rangle = \frac{1}{2}$$ $$P(1) = \text{Tr}(\rho |1\rangle\langle 1|) = \langle 1|\rho|1\rangle = \frac{1}{2}$$

The maximally mixed state gives equal probability for all outcomes, regardless of the basis — it contains no information at all.

Worked Example 4.12: A partially mixed state $\rho = \frac{3}{4}|0\rangle\langle 0| + \frac{1}{4}|1\rangle\langle 1| = \begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}$.

In the Z-basis: $P(0) = 3/4$, $P(1) = 1/4$. In the X-basis: $P(+) = \langle +|\rho|+\rangle = \frac{1}{2}\begin{pmatrix} 1 & 1 \end{pmatrix}\begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}\begin{pmatrix} 1 \\ 1 \end{pmatrix} = \frac{1}{2}(3/4 + 1/4) = \frac{1}{2}$.

The off-diagonal elements of $\rho$ are zero, so the X-basis probabilities are uniform — the state has no coherence between $|0\rangle$ and $|1\rangle$.

# Qiskit: Born rule for mixed states
from qiskit.quantum_info import DensityMatrix
import numpy as np

# Pure state: |+>
rho_pure = DensityMatrix(np.array([[0.5, 0.5], [0.5, 0.5]]))
print("Pure state |+>:")
print(f"  P(0) = {rho_pure.probabilities([0])[0]:.4f}")
print(f"  P(1) = {rho_pure.probabilities([0])[1]:.4f}")

# Mixed state: 3/4 |0><0| + 1/4 |1><1|
rho_mixed = DensityMatrix(np.array([[0.75, 0], [0, 0.25]]))
print("Mixed state (3/4|0><0| + 1/4|1><1|):")
print(f"  P(0) = {rho_mixed.probabilities([0])[0]:.4f}")
print(f"  P(1) = {rho_mixed.probabilities([0])[1]:.4f}")

# Maximally mixed state
rho_max_mixed = DensityMatrix(np.array([[0.5, 0], [0, 0.5]]))
print("Maximally mixed state I/2:")
print(f"  P(0) = {rho_max_mixed.probabilities([0])[0]:.4f}")
print(f"  P(1) = {rho_max_mixed.probabilities([0])[1]:.4f}")

4.3 Projective (von Neumann) Measurements

4.3.1 Definition

A projective measurement is defined by a set of projection operators $\{P_k\}$ acting on the system's Hilbert space $\mathcal{H}$ that satisfy:

  1. Completeness: $\sum_k P_k = I$ (the identity operator).
  2. Orthogonality: $P_k P_j = \delta_{kj} P_k$ (projectors onto mutually orthogonal subspaces).

Each projector $P_k$ corresponds to a measurement outcome labeled by $k$. The probability of outcome $k$ given state $|\psi\rangle$ is:

$$P(k) = \langle \psi | P_k | \psi \rangle$$

For a rank-1 projector $P_k = |a_k\rangle\langle a_k|$ (as in measuring an observable with non-degenerate spectrum), this reduces to $P(k) = |\langle a_k|\psi\rangle|^2$, which is the Born rule.

For degenerate eigenvalues, the projector $P_k$ projects onto the entire eigenspace, and the probability is the total probability weight in that subspace. For example, measuring the total spin squared of a two-electron system yields three outcomes ($S^2 = 0$ or $S^2 = 2\hbar^2$), but the $S^2 = 2\hbar^2$ eigenspace is three-dimensional (the triplet), so the projector has rank 3.

4.3.2 The Collapse Postulate

After a measurement yields outcome $k$, the state of the system collapses to the normalized projection:

$$|\psi\rangle \xrightarrow{\text{outcome } k} \frac{P_k |\psi\rangle}{\sqrt{\langle \psi | P_k | \psi \rangle}}$$

This is the collapse postulate (also called the Lüders rule or projection postulate). It is not derivable from the Schrödinger equation; it is an independent axiom of quantum mechanics.

The collapse postulate has several important consequences:

  1. Irreversibility: The pre-measurement state cannot be recovered from the post-measurement state. Information about the components orthogonal to $P_k$ is irretrievably lost.

  2. Determinism of repeated measurement: A second measurement of the same observable yields the same outcome with certainty. This is because the collapsed state is an eigenstate of the measured observable.

  3. Non-unitarity: Collapse is a non-unitary, non-linear transformation. It cannot be described by a Hamiltonian evolution.

Example: Repeated measurement. If we measure a qubit in state $|+\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$ in the computational basis and obtain $|0\rangle$, a second measurement in the same basis will yield $|0\rangle$ with certainty. The first measurement collapsed the state to $|0\rangle$, and the second measurement simply confirms it.

ASCII Art: Measurement as Projection onto the Bloch Sphere Axes

                    |0⟩ (north pole)
                    |
                    |
        |+⟩ --------+-------- |-⟩
                   /|\
                  / | \
                 /  |  \
                /   |   \
                    |
                    |1⟩ (south pole)

A measurement in the Z-basis projects the state onto either
the |0⟩ or |1⟩ axis. The state |+⟩ (on the equator) collapses
to |0⟩ with probability 1/2 and to |1⟩ with probability 1/2.

State before measurement:      State after measurement:
    (any point on Bloch)    →    (on Z-axis: |0⟩ or |1⟩)

Coherence (off-diagonal      Incoherent mixture:
elements of density            diagonal only
matrix) is destroyed

4.3.3 Worked Example: Three-Outcomes Measurement on a Qutrit

To illustrate projective measurements beyond qubits, consider a three-level system (qutrit) with basis $\{|0\rangle, |1\rangle, |2\rangle\}$ and observable:

$$M = 2|0\rangle\langle 0| + |1\rangle\langle 1| + |1\rangle\langle 2| + |2\rangle\langle 1| + 2|2\rangle\langle 2|$$

Wait — let's use a simpler example. Let the observable be $Z_3 = 2|0\rangle\langle 0| + 0|1\rangle\langle 1| - 2|2\rangle\langle 2|$ (the qutrit Pauli-Z analog) with eigenvalues $2, 0, -2$.

For state $|\psi\rangle = \frac{1}{\sqrt{6}}|0\rangle + \frac{2}{\sqrt{6}}|1\rangle + \frac{1}{\sqrt{6}}|2\rangle$:

$$P(\text{eigenvalue } 2) = |\langle 0|\psi\rangle|^2 = \frac{1}{6}$$ $$P(\text{eigenvalue } 0) = |\langle 1|\psi\rangle|^2 = \frac{4}{6} = \frac{2}{3}$$ $$P(\text{eigenvalue } -2) = |\langle 2|\psi\rangle|^2 = \frac{1}{6}$$

After measuring and obtaining eigenvalue 0 (probability $\frac{2}{3}$), the state collapses to $|1\rangle$.

Try It Yourself: For the state $|\psi\rangle = \frac{1}{2}|0\rangle + \frac{1}{2}|1\rangle + \frac{1}{\sqrt{2}}|2\rangle$, compute the probabilities for each outcome of $Z_3$ and verify they sum to 1. What is the post-measurement state if eigenvalue $-2$ is observed?

4.3.4 Observable Representation

Every projective measurement can be associated with a Hermitian operator (an observable):

$$M = \sum_k m_k P_k$$

where $m_k$ are real numbers (the "values" of the measurement outcomes). The expectation value of $M$ in state $|\psi\rangle$ is:

$$\langle M \rangle = \langle \psi | M | \psi \rangle = \sum_k m_k P(k)$$

The variance is $\langle M^2 \rangle - \langle M \rangle^2$. The standard deviation $\Delta M = \sqrt{\langle M^2 \rangle - \langle M \rangle^2}$ quantifies the intrinsic quantum uncertainty.

Worked Example 4.6: For $|\psi\rangle = \frac{1}{\sqrt{3}}|0\rangle + \sqrt{\frac{2}{3}}|1\rangle$ and the observable $Z$:

$$\langle Z \rangle = \langle \psi|Z|\psi\rangle = |\langle 0|\psi\rangle|^2 \cdot 1 + |\langle 1|\psi\rangle|^2 \cdot (-1) = \frac{1}{3} - \frac{2}{3} = -\frac{1}{3}$$

$$\langle Z^2 \rangle = \langle \psi|Z^2|\psi\rangle = \langle \psi|I|\psi\rangle = 1$$

$$\Delta Z = \sqrt{1 - \frac{1}{9}} = \sqrt{\frac{8}{9}} = \frac{2\sqrt{2}}{3}$$

This is the intrinsic uncertainty of measuring $Z$ on this state — it will not be reduced by better equipment, only by choosing a different state or a different observable.

The Uncertainty Relation. For two observables $A$ and $B$ that do not commute ($[A, B] \neq 0$), the Heisenberg uncertainty relation states:

$$\Delta A \cdot \Delta B \geq \frac{1}{2}|\langle [A, B] \rangle|$$

For $A = X$ and $B = Y$, we have $[X, Y] = 2iZ$, so $\Delta X \cdot \Delta Y \geq |\langle Z \rangle|$. For the state $|0\rangle$, $\langle Z \rangle = 1$, giving $\Delta X \cdot \Delta Y \geq 1$.

4.3.5 Worked Examples: Projective Measurements and Expectation Values

Worked Example 4.13: Let $|\psi\rangle = \frac{1}{\sqrt{5}}|0\rangle + \frac{2}{\sqrt{5}}|1\rangle$. Compute $\langle X \rangle$, $\langle Y \rangle$, $\langle Z \rangle$ and their uncertainties.

$$\langle Z \rangle = \langle\psi|Z|\psi\rangle = |\alpha|^2 - |\beta|^2 = \frac{1}{5} - \frac{4}{5} = -\frac{3}{5}$$

$$\langle Z^2 \rangle = \langle\psi|Z^2|\psi\rangle = \langle\psi|I|\psi\rangle = 1$$

$$\Delta Z = \sqrt{1 - 9/25} = \sqrt{16/25} = \frac{4}{5}$$

For $X$: $\langle X \rangle = \langle\psi|X|\psi\rangle = \alpha^*\beta + \beta^*\alpha = \frac{1}{\sqrt{5}} \cdot \frac{2}{\sqrt{5}} + \frac{2}{\sqrt{5}} \cdot \frac{1}{\sqrt{5}} = \frac{4}{5}$

$\langle X^2 \rangle = 1$, so $\Delta X = \sqrt{1 - 16/25} = \frac{3}{5}$.

Verify uncertainty relation: $\Delta X \cdot \Delta Y \geq |\langle Z \rangle|$. We need $\Delta Y$:

$\langle Y \rangle = i(\alpha\beta^* - \alpha^*\beta) = i(\frac{1}{\sqrt{5}} \cdot \frac{2}{\sqrt{5}} - \frac{1}{\sqrt{5}} \cdot \frac{2}{\sqrt{5}}) = 0$

$\Delta Y = \sqrt{1 - 0} = 1$.

Check: $\Delta X \cdot \Delta Y = \frac{3}{5} \cdot 1 = \frac{3}{5} \geq |\langle Z \rangle| = \frac{3}{5}$. ✓ (Equality holds, so this is a minimum-uncertainty state.)

Worked Example 4.14: Sequential measurements. A qubit starts in $|0\rangle$. We measure $X$ (Hadamard basis), then $Z$ (computational basis).

After the $X$ measurement: - Outcome $+$ with probability $\frac{1}{2}$: state collapses to $|+\rangle$ - Outcome $-$ with probability $\frac{1}{2}$: state collapses to $|-\rangle$

If outcome was $+$: $Z$ measurement gives $P(0) = |\langle 0|+\rangle|^2 = \frac{1}{2}$, $P(1) = \frac{1}{2}$. If outcome was $-$: $Z$ measurement gives $P(0) = |\langle 0|-\rangle|^2 = \frac{1}{2}$, $P(1) = \frac{1}{2}$.

Overall $Z$-measurement probabilities: $P(0) = \frac{1}{2}$, $P(1) = \frac{1}{2}$. The intermediate $X$-measurement has "erased" the information that the qubit was initially in $|0\rangle$.

Worked Example 4.15: Measuring an entangled pair. Consider $|\Phi^+\rangle = \frac{1}{\sqrt{2}}(|00\rangle + |11\rangle)$.

If we measure qubit 1 in the Z-basis: - Outcome 0 (prob 1/2): state collapses to $|00\rangle$. Qubit 2 is in $|0\rangle$. - Outcome 1 (prob 1/2): state collapses to $|11\rangle$. Qubit 2 is in $|1\rangle$.

If we measure qubit 1 in the X-basis instead:

$|\Phi^+\rangle = \frac{1}{\sqrt{2}}(|++\rangle + |--\rangle)$

  • Outcome + (prob 1/2): state collapses to $|++\rangle$. Qubit 2 is in $|+\rangle$.
  • Outcome - (prob 1/2): state collapses to $|--\rangle$. Qubit 2 is in $|-\rangle$.

Notice: the basis we choose for qubit 1 determines whether qubit 2 collapses to a Z-eigenstate or an X-eigenstate. This is the basis-dependence of the "collapse" and is at the heart of the EPR paradox.

# Qiskit: Demonstrating basis-dependent collapse in entangled pairs
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator

simulator = AerSimulator()

# Measure qubit 1 in Z-basis, then qubit 2 in Z-basis
qc_zz = QuantumCircuit(2, 2)
qc_zz.h(0)
qc_zz.cx(0, 1)  # Create |Phi+>
qc_zz.measure([0, 1], [0, 1])
result_zz = simulator.run(qc_zz, shots=10000).result()
print("Z-Z measurement:", result_zz.get_counts())
print("Perfect correlation: 00 and 11 only")

# Measure qubit 1 in X-basis, then qubit 2 in X-basis
qc_xx = QuantumCircuit(2, 2)
qc_xx.h(0)
qc_xx.cx(0, 1)  # Create |Phi+>
qc_xx.h(0)  # Rotate to X-basis for qubit 1
qc_xx.h(1)  # Rotate to X-basis for qubit 2
qc_xx.measure([0, 1], [0, 1])
result_xx = simulator.run(qc_xx, shots=10000).result()
print("\nX-X measurement:", result_xx.get_counts())
print("Perfect correlation in X-basis too: 00 and 11 only")

4.3.6 Complete Sets of Commuting Observables

For a projective measurement, the projectors $\{P_k\}$ correspond to the eigenspaces of an observable $M$. When $M$ has degenerate eigenvalues (eigenspaces of dimension > 1), a single measurement of $M$ does not fully determine the state. We need a complete set of commuting observables (CSCO) — a set of mutually commuting observables whose simultaneous eigenspaces are all one-dimensional.

Example: For a single qubit, $Z$ alone is a CSCO because its eigenvalues ($+1$ and $-1$) are non-degenerate. For two qubits, $Z \otimes Z$ has eigenvalues $+1$ (eigenspace spanned by $|00\rangle, |11\rangle$) and $-1$ (eigenspace spanned by $|01\rangle, |10\rangle$), so it is not a CSCO. But $\{Z \otimes Z, Z \otimes I\}$ (or equivalently $\{Z \otimes Z, I \otimes Z\}$) forms a CSCO: the simultaneous eigenspaces are $|00\rangle, |01\rangle, |10\rangle, |11\rangle$, all one-dimensional.

This is why measuring both qubits in the computational basis is equivalent to measuring the CSCO $\{Z_1, Z_2\}$ — it fully determines the outcome.

4.4 Distinguishability of Quantum States

A fundamental question: given a single copy of a quantum system that is guaranteed to be in one of two known states $|\psi\rangle$ or $|\phi\rangle$, can we determine which one with certainty?

4.4.1 Orthogonal States Are Perfectly Distinguishable

If $|\psi\rangle$ and $|\phi\rangle$ are orthogonal ($\langle \psi|\phi\rangle = 0$), then there exists a measurement that distinguishes them perfectly. Simply choose a basis that includes both states. The outcome corresponding to $|\psi\rangle$ occurs with probability 1 when the state is $|\psi\rangle$ and probability 0 when it is $|\phi\rangle$.

Example: $|0\rangle$ and $|1\rangle$ are orthogonal. A Z-basis measurement distinguishes them with certainty: if the state is $|0\rangle$, we always measure 0; if it is $|1\rangle$, we always measure 1.

4.4.2 Non-Orthogonal States Cannot Be Perfectly Distinguished

Theorem (Non-orthogonal state discrimination). Two pure states $|\psi\rangle$ and $|\phi\rangle$ are perfectly distinguishable by a single projective measurement if and only if they are orthogonal: $\langle \psi | \phi \rangle = 0$.

Proof. If $|\psi\rangle$ and $|\phi\rangle$ are orthogonal, we can choose a measurement basis that includes both states. The outcome corresponding to $|\psi\rangle$ occurs with probability 1 when the state is $|\psi\rangle$ and probability 0 when it is $|\phi\rangle$, and vice versa.

Conversely, suppose there exists a measurement $\{P_k\}$ that perfectly distinguishes them. Then there must be some outcome $k$ that occurs with probability 1 for $|\psi\rangle$ and 0 for $|\phi\rangle$. This means $\langle \psi | P_k | \psi \rangle = 1$ and $\langle \phi | P_k | \phi \rangle = 0$. The first condition implies $P_k |\psi\rangle = |\psi\rangle$ (since $P_k$ is a projector with eigenvalues 0 and 1). The second implies $P_k |\phi\rangle = 0$. Then:

$$\langle \psi | \phi \rangle = \langle \psi | P_k | \phi \rangle = 0$$

Thus orthogonality is necessary. $\square$

Corollary. Non-orthogonal quantum states cannot be distinguished with certainty. This fact underpins quantum cryptography (BB84), where an eavesdropper cannot distinguish the non-orthogonal signal states without introducing detectable disturbances.

4.4.3 Minimum Error Discrimination: The Helstrom Bound

Even when perfect discrimination is impossible, we can ask: what is the best possible probability of correctly identifying the state? Given states $|\psi\rangle$ and $|\phi\rangle$ with prior probabilities $\pi_\psi$ and $\pi_\phi = 1 - \pi_\psi$, the maximum probability of correct identification is given by the Helstrom bound:

$$P_{\text{correct}} = \frac{1}{2}\left(1 + \sqrt{1 - 4\pi_\psi\pi_\phi|\langle\psi|\phi\rangle|^2}\right)$$

For equiprobable states ($\pi_\psi = \pi_\phi = \frac{1}{2}$):

$$P_{\text{correct}} = \frac{1}{2}\left(1 + \sqrt{1 - |\langle\psi|\phi\rangle|^2}\right)$$

Worked Example 4.7: For $|\psi\rangle = |0\rangle$ and $|\phi\rangle = |+\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$, with $\langle\psi|\phi\rangle = \frac{1}{\sqrt{2}}$:

$$P_{\text{correct}} = \frac{1}{2}\left(1 + \sqrt{1 - \frac{1}{2}}\right) = \frac{1}{2}\left(1 + \frac{1}{\sqrt{2}}\right) \approx 0.8536$$

The best possible strategy gives about 85.4% success rate — far below 100%.

Common Misconception: "If we can't distinguish two states perfectly, we just need better measurement devices." No! The indistinguishability of non-orthogonal states is a fundamental law of nature, not an engineering limitation. It follows directly from the linearity of quantum mechanics. No amount of technological improvement can overcome it — it is baked into the mathematics.

# Qiskit: Attempting to distinguish |0> and |+> (non-orthogonal)
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator

def measure_state(state_prep_gate, label):
    """Prepare a state and measure in the Z-basis."""
    qc = QuantumCircuit(1, 1)
    if state_prep_gate == '0':
        pass  # |0> is the default
    elif state_prep_gate == '+':
        qc.h(0)  # |+> = H|0>
    qc.measure(0, 0)
    simulator = AerSimulator()
    job = simulator.run(qc, shots=10000)
    counts = job.result().get_counts()
    print(f"State |{label}> measured in Z-basis: {counts}")

measure_state('0', '0')
measure_state('+', '+')
print("|0> and |+> produce identical statistics in the Z-basis — they are indistinguishable here.")
print("This demonstrates the impossibility of perfect non-orthogonal state discrimination.")

4.5 POVMs: Positive Operator-Valued Measures

Projective measurements are not the most general kind of measurement allowed by quantum mechanics. Positive Operator-Valued Measures (POVMs) provide a broader framework that is essential for quantum information theory.

4.5.1 Definition

A POVM is a set of positive semi-definite operators $\{E_k\}$ on $\mathcal{H}$ satisfying:

$$\sum_k E_k = I$$

Each $E_k$ is called a POVM element or effect. The probability of outcome $k$ given state $|\psi\rangle$ is:

$$P(k) = \langle \psi | E_k | \psi \rangle$$

Unlike projective measurements, POVM elements need not be orthogonal projectors. They need not even be projectors at all — they can be any positive operators summing to the identity.

The relationship between POVMs and projective measurements is given by Naimark's dilation theorem: every POVM on $\mathcal{H}$ can be realized as a projective measurement on $\mathcal{H} \otimes \mathcal{H}_{\text{ancilla}}$ for some ancilla system. This means POVMs are not fundamentally more powerful than projective measurements — they are just more convenient for describing certain scenarios, particularly when we don't care about the post-measurement state.

4.5.2 Why POVMs? The Advantage of Flexibility

POVMs are useful when we do not care about the post-measurement state — only the outcome probabilities matter. This is common in quantum state discrimination, where we want to maximize the probability of correctly identifying a state.

Key advantages of POVMs over projective measurements:

  1. More outcomes than dimension: A POVM on $\mathbb{C}^2$ can have any number of outcomes, not just 2. This is impossible for projective measurements, which are limited by the dimension of the Hilbert space.

  2. Better discrimination: For state discrimination tasks, POVMs can achieve higher success rates than any projective measurement.

  3. Unambiguous discrimination: POVMs enable a trade-off between certainty and completeness — we can sometimes identify a state with certainty, at the cost of sometimes getting no answer.

4.5.3 Example: Unambiguous State Discrimination

Given a single copy of a qubit that is either $|0\rangle$ or $|+\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$, we can construct a POVM that sometimes identifies the state with certainty and sometimes yields an inconclusive result — but never makes a mistake.

The POVM elements are:

$$E_0 = a|1\rangle\langle 1|, \quad E_+ = a|-\rangle\langle -|, \quad E_? = I - E_0 - E_+$$

where $a = \frac{2}{1 + 1/\sqrt{2}}$ ensures positivity. If outcome 0 occurs, the state was definitely $|+\rangle$ (since $\langle 0|E_0|0\rangle = 0$). If outcome + occurs, the state was definitely $|0\rangle$. If outcome ? occurs, we learn nothing.

Full derivation. We need $E_0, E_+, E_? \geq 0$ and $E_0 + E_+ + E_? = I$. Let $E_0 = a|1\rangle\langle 1|$ and $E_+ = a|-\rangle\langle -|$. For these to yield unambiguous discrimination:

  • $E_0$ should have zero probability when the state is $|0\rangle$: $\langle 0|E_0|0\rangle = 0$ ✓ (since $|1\rangle \perp |0\rangle$).
  • $E_+$ should have zero probability when the state is $|+\rangle$: $\langle +|E_+|+\rangle = 0$ ✓ (since $|-\rangle \perp |+\rangle$).

The parameter $a$ must satisfy $E_? = I - E_0 - E_+ \geq 0$. The maximum $a$ occurs when $E_?$ has a zero eigenvalue (boundary of positivity):

$$a_{\max} = \frac{1}{\||\psi\rangle\langle 1| + |\phi\rangle\langle -|\|} = \frac{2}{1 + 1/\sqrt{2}} \approx 1.172$$

Wait, that's not right. Let us redo this carefully. We need $I - a|1\rangle\langle 1| - a|-\rangle\langle -| \geq 0$. In the computational basis:

$$|1\rangle\langle 1| = \begin{pmatrix} 0 & 0 \\ 0 & 1 \end{pmatrix}, \quad |-\rangle\langle -| = \frac{1}{2}\begin{pmatrix} 1 & -1 \\ -1 & 1 \end{pmatrix}$$

$$I - a|1\rangle\langle 1| - a|-\rangle\langle -| = \begin{pmatrix} 1 - a/2 & a/2 \\ a/2 & 1 - a - a/2 \end{pmatrix} = \begin{pmatrix} 1 - a/2 & a/2 \\ a/2 & 1 - 3a/2 \end{pmatrix}$$

For this to be positive semi-definite, both eigenvalues must be $\geq 0$, and the determinant must be $\geq 0$:

$$(1 - a/2)(1 - 3a/2) - a^2/4 \geq 0$$ $$1 - 3a/2 - a/2 + 3a^2/4 - a^2/4 \geq 0$$ $$1 - 2a + a^2/2 \geq 0$$

Setting this to zero: $a^2/2 - 2a + 1 = 0$, giving $a = 2 \pm \sqrt{2}$. Since $a \leq 1$ (for $1 - a/2 \geq 0$), we take $a = 2 - \sqrt{2} \approx 0.586$.

The probability of an inconclusive result when the input is $|0\rangle$ (equiprobable) is:

$$P(?) = \frac{1}{2}\langle 0|E_?|0\rangle + \frac{1}{2}\langle +|E_?|+\rangle$$

This probability is the price we pay for certainty — we sometimes get no answer, but we never get a wrong answer.

4.5.4 POVMs for Optimal Minimum-Error Discrimination

For minimum-error discrimination (maximizing the probability of a correct identification, allowing errors), the optimal measurement is a Helstrom measurement. For two pure states $|\psi\rangle$ and $|\phi\rangle$ with prior probabilities $\pi_\psi$ and $\pi_\phi$:

The optimal POVM is the projective measurement onto the positive and negative eigenspaces of the operator $\Lambda = \pi_\psi|\psi\rangle\langle\psi| - \pi_\phi|\phi\rangle\langle\phi|$.

Worked Example 4.8: For $|\psi\rangle = |0\rangle$ and $|\phi\rangle = |+\rangle$ with $\pi_\psi = \pi_\phi = \frac{1}{2}$:

$$\Lambda = \frac{1}{2}|0\rangle\langle 0| - \frac{1}{2}|+\rangle\langle +| = \frac{1}{2}\begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix} - \frac{1}{4}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix} = \begin{pmatrix} 1/4 & -1/4 \\ -1/4 & -1/4 \end{pmatrix}$$

The eigenvalues of $\Lambda$ are $\frac{1}{2\sqrt{2}}$ and $-\frac{1}{2\sqrt{2}}$. The success probability is $\frac{1}{2}(1 + \frac{1}{\sqrt{2}}) \approx 0.854$, matching the Helstrom bound.

# Qiskit: Helstrom measurement for state discrimination
import numpy as np
from qiskit.quantum_info import Statevector, Operator

# Define states
psi = Statevector([1, 0])        # |0>
phi = Statevector([1/np.sqrt(2), 1/np.sqrt(2)])  # |+>

# Compute overlap
overlap = np.abs(psi.inner(phi))
print(f"|<psi|phi>| = {overlap:.4f}")

# Helstrom bound for equiprobable states
p_correct = 0.5 * (1 + np.sqrt(1 - overlap**2))
print(f"Helstrom bound: P(correct) = {p_correct:.4f}")
print(f"This is better than random guessing (0.5) but less than perfect (1.0)")

4.5.5 Naimark's Dilation Theorem

Every POVM can be realized as a projective measurement on a larger Hilbert space (system + ancilla). This is Naimark's dilation theorem and it means POVMs are not fundamentally more powerful than projective measurements — they are just more convenient for describing certain scenarios.

Constructive version. Given a POVM $\{E_k\}$ on $\mathcal{H}_S$, we can construct an ancilla space $\mathcal{H}_A$ with orthonormal basis $\{|a_k\rangle\}$ and a unitary $U$ on $\mathcal{H}_S \otimes \mathcal{H}_A$ such that:

$$E_k = \langle a_k| U^\dagger P_k^{(S)} U |a_0\rangle$$

where $P_k^{(S)}$ is a projector on $\mathcal{H}_S$ and $|a_0\rangle$ is the initial ancilla state. The procedure is:

  1. Prepare the ancilla in $|a_0\rangle$
  2. Apply the unitary $U$ to the combined system
  3. Measure the ancilla in the $\{|a_k\rangle\}$ basis
  4. The system measurement outcome probabilities reproduce the POVM

This provides a concrete physical implementation of any POVM and shows that the measurement formalism is fundamentally self-consistent.

4.6 The No-Cloning Theorem

4.6.1 Statement and Proof

No-Cloning Theorem (Wootters & Zurek, 1982; Dieks, 1982). There is no unitary operator $U$ that can copy an arbitrary unknown quantum state. That is, there is no $U$ such that for all $|\psi\rangle$:

$$U(|\psi\rangle \otimes |0\rangle) = |\psi\rangle \otimes |\psi\rangle$$

Proof (by inner product preservation). Suppose such a $U$ exists. Then for any two states $|\psi\rangle$ and $|\phi\rangle$:

$$U(|\psi\rangle \otimes |0\rangle) = |\psi\rangle \otimes |\psi\rangle$$ $$U(|\phi\rangle \otimes |0\rangle) = |\phi\rangle \otimes |\phi\rangle$$

Take the inner product of both sides. Since $U$ is unitary, it preserves inner products:

$$\langle \psi \otimes 0 | \phi \otimes 0 \rangle = \langle \psi \otimes \psi | \phi \otimes \phi \rangle$$

The left side is $\langle \psi | \phi \rangle \cdot \langle 0 | 0 \rangle = \langle \psi | \phi \rangle$. The right side is $\langle \psi | \phi \rangle \cdot \langle \psi | \phi \rangle = \langle \psi | \phi \rangle^2$.

Thus $\langle \psi | \phi \rangle = \langle \psi | \phi \rangle^2$, which implies $\langle \psi | \phi \rangle \in \{0, 1\}$. This means $|\psi\rangle$ and $|\phi\rangle$ must be either identical or orthogonal. But the theorem claims the cloner works for arbitrary states, including non-orthogonal ones. Contradiction. $\square$

Alternative proof (by linearity). Suppose $U(|\psi\rangle \otimes |0\rangle) = |\psi\rangle \otimes |\psi\rangle$ for all $|\psi\rangle$. Then for $|\psi\rangle = \alpha|0\rangle + \beta|1\rangle$:

$$U((\alpha|0\rangle + \beta|1\rangle) \otimes |0\rangle) = (\alpha|0\rangle + \beta|1\rangle) \otimes (\alpha|0\rangle + \beta|1\rangle)$$

But by linearity:

$$U(\alpha|0\rangle \otimes |0\rangle + \beta|1\rangle \otimes |0\rangle) = \alpha \cdot U(|00\rangle) + \beta \cdot U(|10\rangle) = \alpha|0\rangle|0\rangle + \beta|1\rangle|1\rangle$$

These are different (the RHS contains cross-terms $|01\rangle$ and $|10\rangle$ absent from the linear version), so linearity is violated. $\square$

4.6.2 Implications

The no-cloning theorem is the bedrock of quantum cryptography. An eavesdropper cannot copy the quantum signals traveling between Alice and Bob without being detected. It also means that quantum error correction cannot work by simply making backup copies of qubits — more sophisticated techniques (encoding into entangled states) are required.

Why this matters for quantum computing:

  1. Quantum cryptography: The security of BB84 and other QKD protocols rests on the impossibility of cloning. An eavesdropper who intercepts a qubit cannot make a copy to study later — measuring the original disturbs it irreversibly.

  2. Quantum error correction: Classical error correction uses redundancy (copying bits). Quantum error correction must use entanglement instead. This is why quantum error-correcting codes encode one logical qubit into many physical qubits in a way that allows error detection and correction without ever copying the unknown state.

  3. Quantum communication: The no-cloning theorem limits the range of quantum communication. A quantum signal cannot be amplified (copied) as a classical signal can. This is why quantum repeaters (which use entanglement swapping rather than signal amplification) are needed for long-distance quantum networks.

  4. Quantum advantage: The inability to copy quantum information is a feature, not a bug. It is what makes quantum key distribution unconditionally secure and what prevents certain classical attacks on quantum protocols.

Recurring Theme — Information Is Physical: The no-cloning theorem illustrates that quantum information obeys fundamentally different rules than classical information. You cannot copy it, you cannot read it without disturbing it, and you cannot broadcast it. These constraints are not technological limitations — they are laws of nature.

Common Misconception: "Quantum computers can copy qubits using CNOT." The CNOT gate maps $|0\rangle|0\rangle \to |0\rangle|0\rangle$ and $|1\rangle|0\rangle \to |1\rangle|1\rangle$, which looks like copying. But for a superposition $|\psi\rangle = \alpha|0\rangle + \beta|1\rangle$:

$\text{CNOT}(|\psi\rangle \otimes |0\rangle) = \text{CNOT}(\alpha|00\rangle + \beta|10\rangle) = \alpha|00\rangle + \beta|11\rangle$

This is an entangled state, NOT $\alpha^2|00\rangle + \alpha\beta|01\rangle + \alpha\beta|10\rangle + \beta^2|11\rangle = |\psi\rangle \otimes |\psi\rangle$. The CNOT creates entanglement, not a copy. You cannot extract two copies of $|\psi\rangle$ from the entangled state $\alpha|00\rangle + \beta|11\rangle$.

4.6.3 The No-Deleting Theorem

A natural complement to the no-cloning theorem: there is no unitary $U$ such that $U(|\psi\rangle \otimes |\psi\rangle) = |\psi\rangle \otimes |0\rangle$ for all $|\psi\rangle$.

Proof sketch: If such a $U$ existed, then combined with a hypothetical cloner, we could have $U_{\text{clone}} U_{\text{delete}}(|\psi\rangle \otimes |0\rangle) = U_{\text{clone}}(|\psi\rangle \otimes |\psi\rangle)$ — but this would allow information to be created, violating the unitarity of quantum mechanics. More directly: take the inner product $\langle\psi \otimes \psi|\phi \otimes \phi\rangle = \langle\psi|\phi\rangle^2$ on the input and $\langle\psi \otimes 0|\phi \otimes 0\rangle = \langle\psi|\phi\rangle$ on the output. These are equal only if $\langle\psi|\phi\rangle = 0$ or $\langle\psi|\phi\rangle = 1$, contradicting universality.

Together, the no-cloning and no-deleting theorems establish a conservation law for quantum information: you can neither create nor destroy copies of an unknown quantum state.

4.7 Quantum State Tomography

If measurement destroys the state, how can we ever determine what the state was? The answer is quantum state tomography: by preparing many identical copies of a state and performing different measurements on each copy, we can reconstruct the density matrix.

4.7.1 Single-Qubit Tomography

A single-qubit state can be written as:

$$\rho = \frac{1}{2}(I + x\sigma_x + y\sigma_y + z\sigma_z) = \frac{1}{2}\begin{pmatrix} 1+z & x-iy \\ x+iy & 1-z \end{pmatrix}$$

where $\vec{r} = (x, y, z)$ is the Bloch vector with $|\vec{r}| \leq 1$. The coefficients are expectation values:

$$x = \langle \sigma_x \rangle = \text{Tr}(\rho \sigma_x)$$ $$y = \langle \sigma_y \rangle = \text{Tr}(\rho \sigma_y)$$ $$z = \langle \sigma_z \rangle = \text{Tr}(\rho \sigma_z)$$

To estimate these, we measure in the $X$, $Y$, and $Z$ bases and compute:

$$\langle \sigma_z \rangle = P(0) - P(1) \quad \text{(measure in Z-basis)}$$ $$\langle \sigma_x \rangle = P(+) - P(-) \quad \text{(measure in X-basis)}$$ $$\langle \sigma_y \rangle = P(+i) - P(-i) \quad \text{(measure in Y-basis)}$$

Why three bases? A qubit has three real parameters (two for the Bloch sphere angles and one for the purity/trace). Each measurement in a basis gives one real number (the expectation value of the corresponding Pauli). Three measurements give exactly three parameters — a complete characterization.

Worked Example 4.9: State $|\psi\rangle = \cos(\theta/2)|0\rangle + e^{i\phi}\sin(\theta/2)|1\rangle$ for $\theta = \pi/4$, $\phi = \pi/3$.

Theoretical values: $$z = \cos\theta = \cos(\pi/4) = \frac{1}{\sqrt{2}}$$ $$x = \sin\theta\cos\phi = \sin(\pi/4)\cos(\pi/3) = \frac{1}{\sqrt{2}} \cdot \frac{1}{2} = \frac{1}{2\sqrt{2}}$$ $$y = \sin\theta\sin\phi = \sin(\pi/4)\sin(\pi/3) = \frac{1}{\sqrt{2}} \cdot \frac{\sqrt{3}}{2} = \frac{\sqrt{3}}{2\sqrt{2}}$$

Bloch vector length: $\sqrt{x^2 + y^2 + z^2} = \sqrt{\frac{1}{8} + \frac{3}{8} + \frac{1}{2}} = 1$ (pure state).

# Qiskit: Single-qubit state tomography
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
import numpy as np

def tomography_1q(state_prep_circuit):
    """Perform single-qubit tomography on a prepared state."""
    simulator = AerSimulator()
    results = {}

    # Z-basis measurement (computational basis)
    qc_z = state_prep_circuit.copy()
    qc_z.measure_all()
    counts_z = simulator.run(qc_z, shots=10000).result().get_counts()
    p0_z = counts_z.get('0', 0) / 10000
    p1_z = counts_z.get('1', 0) / 10000
    results['<Z>'] = p0_z - p1_z

    # X-basis measurement (Hadamard then Z)
    qc_x = state_prep_circuit.copy()
    qc_x.h(0)
    qc_x.measure_all()
    counts_x = simulator.run(qc_x, shots=10000).result().get_counts()
    p0_x = counts_x.get('0', 0) / 10000
    p1_x = counts_x.get('1', 0) / 10000
    results['<X>'] = p0_x - p1_x

    # Y-basis measurement (S† then H then Z)
    qc_y = state_prep_circuit.copy()
    qc_y.sdg(0)
    qc_y.h(0)
    qc_y.measure_all()
    counts_y = simulator.run(qc_y, shots=10000).result().get_counts()
    p0_y = counts_y.get('0', 0) / 10000
    p1_y = counts_y.get('1', 0) / 10000
    results['<Y>'] = p0_y - p1_y

    return results

# Prepare a state: rotate |0> by pi/4 around Y-axis
qc_prep = QuantumCircuit(1, 0)
qc_prep.ry(np.pi/4, 0)

tomo = tomography_1q(qc_prep)
print("Tomography results:")
for obs, val in tomo.items():
    print(f"  {obs} = {val:.4f}")

# Bloch vector length should be ~1 for a pure state
r = np.sqrt(tomo['<X>']**2 + tomo['<Y>']**2 + tomo['<Z>']**2)
print(f"Bloch vector length |r| = {r:.4f} (should be ~1 for pure state)")

4.7.2 Maximum Likelihood Estimation

In practice, finite sampling introduces statistical noise. The raw estimates may produce a Bloch vector with $|\vec{r}| > 1$, which is unphysical. Maximum likelihood estimation (MLE) finds the physical density matrix (positive semi-definite, unit trace) that maximizes the likelihood of the observed data.

The MLE approach parameterizes the density matrix as:

$$\rho = \frac{T^\dagger T}{\text{Tr}(T^\dagger T)}$$

where $T$ is a lower-triangular matrix. This automatically ensures $\rho \geq 0$ and $\text{Tr}(\rho) = 1$. The likelihood function is:

$$\mathcal{L}(\rho) = \prod_k P(k|\rho)^{n_k} = \prod_k \langle \psi_k | \rho | \psi_k \rangle^{n_k}$$

where $n_k$ is the number of times outcome $k$ was observed. Maximizing $\mathcal{L}$ (or equivalently, minimizing $-\log \mathcal{L}$) over the parameters of $T$ gives the MLE estimate.

Qiskit's qiskit_experiments library provides built-in tomography routines that handle MLE automatically.

4.7.3 Two-Qubit Tomography

For two qubits, the state space is 15-dimensional (a $4 \times 4$ density matrix with 15 independent real parameters after accounting for unit trace). The tomography requires measuring in 9 combinations of single-qubit bases (XX, XY, XZ, YX, YY, YZ, ZX, ZY, ZZ), each requiring many shots. The total number of measurements scales as $O(4^n)$ for $n$ qubits, making tomography exponentially expensive — another manifestation of the exponential size of the quantum state space.

# Qiskit: Two-qubit tomography (partial — measuring correlations)
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
import numpy as np

def two_qubit_correlations(state_prep):
    """Measure <Z⊗Z>, <X⊗X>, <Z⊗I>, <I⊗Z> for a two-qubit state."""
    simulator = AerSimulator()
    results = {}

    # <Z⊗Z>: measure both in Z basis
    qc_zz = state_prep.copy()
    qc_zz.measure_all()
    counts = simulator.run(qc_zz, shots=10000).result().get_counts()
    p00 = counts.get('00', 0) / 10000
    p01 = counts.get('01', 0) / 10000
    p10 = counts.get('10', 0) / 10000
    p11 = counts.get('11', 0) / 10000
    results['<ZZ>'] = p00 - p01 - p10 + p11

    # <X⊗X>: measure both in X basis
    qc_xx = state_prep.copy()
    qc_xx.h([0, 1])
    qc_xx.measure_all()
    counts = simulator.run(qc_xx, shots=10000).result().get_counts()
    p00 = counts.get('00', 0) / 10000
    p01 = counts.get('01', 0) / 10000
    p10 = counts.get('10', 0) / 10000
    p11 = counts.get('11', 0) / 10000
    results['<XX>'] = p00 - p01 - p10 + p11

    return results

# Bell state |Phi+>
qc_bell = QuantumCircuit(2)
qc_bell.h(0)
qc_bell.cx(0, 1)

corr = two_qubit_correlations(qc_bell)
print("Bell state correlations:")
for key, val in corr.items():
    print(f"  {key} = {val:.4f}")
print("Expected: <ZZ> = 1, <XX> = 1 (perfect correlations)")

4.7.4 Fidelity and Distance Measures

After performing tomography, we need to quantify how close the reconstructed state $\sigma$ is to the theoretical state $\rho$. The most common measure is the fidelity:

$$F(\rho, \sigma) = \left(\text{Tr}\sqrt{\sqrt{\rho}\sigma\sqrt{\rho}}\right)^2$$

For pure states $\rho = |\psi\rangle\langle\psi|$ and general $\sigma$, this simplifies to:

$$F(|\psi\rangle, \sigma) = \langle\psi|\sigma|\psi\rangle$$

And for two pure states:

$$F(|\psi\rangle, |\phi\rangle) = |\langle\psi|\phi\rangle|^2$$

Worked Example 4.16: Compute the fidelity between $|\psi\rangle = |0\rangle$ and $\sigma = \frac{3}{4}|0\rangle\langle 0| + \frac{1}{4}|1\rangle\langle 1|$.

$$F = \langle 0|\sigma|0\rangle = \frac{3}{4}$$

Worked Example 4.17: Compute the fidelity between $|\psi\rangle = |+\rangle$ and $|\phi\rangle = |0\rangle$.

$$F = |\langle +|0\rangle|^2 = \frac{1}{2}$$

These states are 50% overlapping — neither orthogonal nor identical.

The trace distance is another important measure:

$$T(\rho, \sigma) = \frac{1}{2}\|\rho - \sigma\|_1 = \frac{1}{2}\text{Tr}|\rho - \sigma|$$

where $|A| = \sqrt{A^\dagger A}$. The trace distance has an operational interpretation: it is the maximum probability of distinguishing $\rho$ from $\sigma$ using any measurement.

Relationship between fidelity and trace distance: For pure states, $T(|\psi\rangle, |\phi\rangle) = \sqrt{1 - |\langle\psi|\phi\rangle|^2} = \sqrt{1 - F}$. In general, the Fuchs-van de Graaf inequalities relate them:

$$1 - \sqrt{F(\rho, \sigma)} \leq T(\rho, \sigma) \leq \sqrt{1 - F(\rho, \sigma)}$$

# Qiskit: Computing fidelity and trace distance
from qiskit.quantum_info import Statevector, DensityMatrix, state_fidelity
import numpy as np

# Pure state fidelity
psi = Statevector([1, 0])  # |0>
phi = Statevector([1/np.sqrt(2), 1/np.sqrt(2)])  # |+>

F = abs(psi.inner(phi))**2
print(f"Fidelity |<0|+>|^2 = {F:.4f}")

# Trace distance between pure states
T = np.sqrt(1 - F)
print(f"Trace distance = {T:.4f}")

# Fidelity with mixed state
rho_mixed = DensityMatrix(np.array([[0.75, 0], [0, 0.25]]))
F_mixed = rho_mixed.data[0, 0]  # <0|rho|0> = 0.75
print(f"\nFidelity F(|0>, rho_mixed) = {F_mixed:.4f}")

# Compare with maximally mixed state
print(f"Fidelity F(|0>, I/2) = 0.5000")  # <0|I/2|0> = 1/2

4.7.5 Process Tomography

Just as state tomography reconstructs the density matrix, process tomography (or quantum process tomography) reconstructs the quantum channel $\mathcal{E}$ that describes how a gate transforms states. A quantum channel maps input states to output states:

$$\rho_{\text{out}} = \mathcal{E}(\rho_{\text{in}})$$

The most general representation uses the chi matrix (or process matrix):

$$\mathcal{E}(\rho) = \sum_{m,n} \chi_{mn} E_m \rho E_n^\dagger$$

where $\{E_m\}$ is a basis for the space of operators (typically the Pauli basis for one qubit: $\{I, X, Y, Z\}$) and $\chi$ is a $4 \times 4$ positive semi-definite matrix.

Process tomography requires: 1. Preparing a complete set of input states (for one qubit: $|0\rangle, |1\rangle, |+\rangle, |+\rangle_y$) 2. Applying the gate to each input 3. Performing state tomography on each output 4. Reconstructing $\chi$ from the input-output pairs

For a single-qubit gate, this requires $4 \times 3 = 12$ measurement settings (4 inputs × 3 measurement bases each). For $n$ qubits, the chi matrix is $4^n \times 4^n$, requiring $4^n \times 3^n$ measurement settings — exponentially expensive, just like state tomography.

# Qiskit: Process tomography for the Hadamard gate
from qiskit import QuantumCircuit
from qiskit.quantum_info import Statevector, Operator
import numpy as np

def process_tomography_1q(gate_circuit, label=""):
    """Perform process tomography on a single-qubit gate."""
    # Prepare 4 input states
    input_circuits = {}

    # |0> state
    input_circuits['|0>'] = QuantumCircuit(1)

    # |1> state
    qc1 = QuantumCircuit(1)
    qc1.x(0)
    input_circuits['|1>'] = qc1

    # |+> state
    qcp = QuantumCircuit(1)
    qcp.h(0)
    input_circuits['|+>'] = qcp

    # |+i> state  
    qci = QuantumCircuit(1)
    qci.h(0)
    qci.s(0)
    input_circuits['|+i>'] = qci

    print(f"Process tomography for {label}:")
    for name, prep in input_circuits.items():
        qc = prep.copy()
        qc.compose(gate_circuit, inplace=True)
        sv = Statevector.from_instruction(qc)
        print(f"  Input {name} → Output state: {np.round(sv.data, 4)}")

    # Theoretical output for H gate
    H = np.array([[1, 1], [1, -1]]) / np.sqrt(2)
    print(f"\nTheoretical H|0> = {np.round(H @ [1, 0], 4)}")
    print(f"Theoretical H|1> = {np.round(H @ [0, 1], 4)}")

# Process tomography of the Hadamard gate
h_gate = QuantumCircuit(1)
h_gate.h(0)
process_tomography_1q(h_gate, "Hadamard")

4.8 Measurement in Quantum Circuits

In the circuit model, measurement is represented by a meter symbol:

ASCII Art: Quantum Circuit with Measurement

q0: ──[H]──[M]────
            │
c0: ────────╨──────   (classical register stores result)

Measurement is typically performed at the end of a quantum circuit, collapsing the final state and producing classical bits. However, mid-circuit measurements are also possible and are essential for quantum error correction, teleportation, and feed-forward operations.

4.8.1 Mid-Circuit Measurements and Feed-Forward

Mid-circuit measurements allow classical information extracted during computation to influence subsequent quantum operations. This is crucial for:

  1. Quantum teleportation: Alice's measurement outcome determines Bob's correction
  2. Quantum error correction: Syndrome measurements guide correction operations
  3. Measurement-based quantum computing: The entire computation is driven by measurement outcomes
# Qiskit: Mid-circuit measurement and conditional operations
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator

qc = QuantumCircuit(2, 2)
qc.h(0)
qc.measure(0, 0)           # Mid-circuit measurement
# Conditionally apply X gate based on measurement result
qc.x(1).c_if(0, 1)         # If bit 0 is 1, flip qubit 1
qc.measure(1, 1)

simulator = AerSimulator()
job = simulator.run(qc, shots=10000)
counts = job.result().get_counts()
print("Mid-circuit measurement results:", counts)

4.8.2 Measurement and Classical Control Flow

ASCII Art: Teleportation Circuit with Measurements

q0 (|ψ>): ──[H]──■──────────[M]──────────────────
                   │          │
q1 (|0>):  ───────[X]──[H]──[M]──────────────────
                                        │  │
q2 (|0>):  ───────────────────[X]────[Z]──[M]──
                                       ↑  ↑
                              b₀ ──────┘  │
                              b₁ ─────────┘

Alice's Bell measurement collapses the state and produces
two classical bits b₀, b₁. Bob applies X^b₀ and Z^b₁
to recover |ψ⟩ on q2.

The circuit model with mid-circuit measurement and classical control is strictly more powerful than the unitary-only model (though the additional power is subtle and mainly relevant for distributed protocols).

4.8.3 Deferred Measurement Principle

An important theorem: any quantum circuit with mid-circuit measurements can be transformed into an equivalent circuit where all measurements are at the end, at the cost of introducing ancilla qubits. This is the deferred measurement principle.

Specifically, if a measurement of qubit $q$ with outcome stored in classical bit $c$ controls a gate $U$ on qubit $q'$:

  1. Replace the measurement with a CNOT from $q$ to a fresh ancilla $a$
  2. Replace the classically controlled $U$ with a controlled-$U$ with $a$ as control

This transformation is used in theoretical proofs and in understanding the relationship between measurement-based and circuit-based quantum computation.

Common Misconception: "Mid-circuit measurements are always necessary." By the deferred measurement principle, any circuit with mid-circuit measurements can be reformulated with only end-of-circuit measurements (at the cost of extra ancillas). However, mid-circuit measurements are often practically essential because the deferred version requires more qubits, which are a scarce resource on current hardware.

4.9 The Measurement Problem (Interpretational)

We have presented measurement as an operational recipe: Born rule gives probabilities, collapse gives the post-measurement state. But why does collapse happen? What constitutes a "measurement"? Where is the boundary between quantum system and classical apparatus?

These questions constitute the measurement problem and have spawned numerous interpretations:

  • Copenhagen interpretation: The wavefunction is a tool for calculating probabilities; collapse is real but not further explained. The act of observation plays a fundamental role. This is the "shut up and calculate" approach, and it's the operational framework that most working quantum information scientists adopt.

  • Many-Worlds (Everett): There is no collapse. The universe splits into branches, one for each outcome, and we experience only one branch. The appearance of collapse arises from decoherence — each branch becomes effectively independent. The Born rule is derived from the branch weights.

  • Decoherence: Interaction with the environment effectively selects a preferred basis (the "pointer basis") and suppresses interference between branches, explaining why the world appears classical without requiring a fundamental collapse. Decoherence is not an interpretation but a physical process that any interpretation must account for.

  • QBism (Quantum Bayesianism): Quantum states are degrees of belief held by an agent, not physical realities. Measurement is Bayesian updating of the agent's beliefs. The wavefunction doesn't collapse — the agent's knowledge updates.

  • Objective collapse models (GRW, Penrose): Collapse is a real physical process triggered by some mechanism (spontaneous localization, gravitational effects). These models modify the Schrödinger equation and make testable predictions that differ from standard quantum mechanics.

For the practicing quantum computer scientist, the operational rules suffice. But the interpretational questions remind us that quantum mechanics, for all its predictive power, remains philosophically unsettled.

Recurring Theme — The Classical-Quantum Interface: Every quantum computation ends at a classical readout. Understanding measurement is understanding how quantum advantage translates into classical answers we can use. The measurement problem is not just a philosophical curiosity — it defines the boundary at which quantum effects become classical information.

ASCII Art: The Measurement Problem — Where Does Collapse Happen?

    Quantum system          Apparatus         Observer
    ┌───────────┐        ┌──────────┐      ┌──────────┐
    │ |ψ⟩ = α|0⟩│───────>│ detector  │─────>│ conscious│
    │    + β|1⟩ │        │ (pointer) │      │  observer │
    └───────────┘        └──────────┘      └──────────┘
         ↑                     ↑                  ↑
    Unitary            Decoherence?          Consciousness?
    evolution          (environment           (von Neumann
    (Schrödinger)      entanglement)          chain)

Where does the transition from quantum superposition
to classical outcome occur? This is the measurement problem.

4.9.1 Measurement in Different Qubit Technologies

Different quantum computing platforms implement measurement differently:

Superconducting qubits (IBM, Google): Measurement is performed by coupling the qubit to a microwave resonator and reading out the transmitted/reflected signal. The qubit state shifts the resonator frequency, producing a different response for $|0\rangle$ and $|1\rangle$. Typical measurement times are 0.5-2 μs with fidelities of 95-99%.

Trapped ions (IonQ, Quantinuum): Measurement uses fluorescence — a laser illuminates the ion, and it fluoresces (scatters photons) only if in one of the two states. A photomultiplier tube counts photons to determine the state. Measurement fidelities of 99.5-99.9% are typical.

Photonic qubits: Measurement uses single-photon detectors (avalanche photodiodes, superconducting nanowire detectors). The detection of a photon in a particular mode corresponds to a measurement outcome. Photon number resolution is possible with transition-edge sensors.

Neutral atoms (Atom Computing, QuEra): Similar to trapped ions — fluorescence detection with global illumination and site-resolved imaging.

ASCII Art: Measurement Technologies Comparison

┌──────────────────────┬──────────────┬──────────────┬─────────────┐
│ Platform              │ Method       │ Time        │ Fidelity    │
├──────────────────────┼──────────────┼──────────────┼─────────────┤
│ Superconducting       │ Resonator    │ 0.5-2 μs    │ 95-99%      │
│ Trapped ions          │ Fluorescence │ 10-100 μs   │ 99.5-99.9%  │
│ Photonic              │ SP detectors │ ~ns         │ 90-99%      │
│ Neutral atoms         │ Fluorescence │ 1-10 μs     │ 97-99.5%    │
└──────────────────────┴──────────────┴──────────────┴─────────────┘

Recurring Theme — We're at the Beginning: Current measurement fidelities of 95-99% are far from ideal. Each measurement error propagates through the computation. Until measurement fidelities improve or error correction compensates, measurement noise will limit the depth of quantum circuits that can produce useful results.

4.9.2 Weak Measurement and Quantum Trajectories

The standard projective measurement model is an idealization — it assumes measurement is instantaneous and complete. In reality, measurement is a continuous process. Weak measurement describes the regime where the measurement is partial, extracting only a small amount of information per unit time.

A weak measurement of observable $A$ with strength $\lambda$ produces outcome $m$ with probability:

$$P(m) = \langle\psi|e^{-\lambda(m-A)^2}|\psi\rangle / Z$$

where $Z$ is a normalization constant. As $\lambda \to \infty$, this approaches a projective measurement. As $\lambda \to 0$, the measurement extracts vanishingly little information per outcome.

The post-measurement state after a weak measurement is:

$$|\psi\rangle \to \frac{e^{-\lambda(m-A)^2/2}|\psi\rangle}{\sqrt{\langle\psi|e^{-\lambda(m-A)^2}|\psi\rangle}}$$

Weak measurements are important for: 1. Quantum trajectories: Continuously monitoring a quantum system produces a stochastic evolution (a "quantum trajectory") that interpolates between unitary evolution (no measurement) and collapse (strong measurement). 2. Quantum feedback control: Partial measurement results are used to adjust the system in real time, stabilizing desired states or correcting errors before they become catastrophic. 3. Quantum state smoothing: Using both past and future measurement records to reconstruct the state more accurately than filtering alone.

# Qiskit: Simulating a weak measurement effect
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
from qiskit.quantum_info import Statevector, DensityMatrix
import numpy as np

# Prepare |+> state and apply a weak Z measurement
# A weak Z measurement with strength lambda can be modeled as:
# rho -> (1-p)|+><+| + p|0><0|  (for outcome 0)
# or    (1-p)|+><+| + p|1><1|  (for outcome 1)
# where p is the measurement strength

def weak_z_measurement(rho, strength):
    """Apply a weak Z measurement to density matrix rho."""
    P0 = np.array([[1, 0], [0, 0]])
    P1 = np.array([[0, 0], [0, 1]])

    # Weak measurement Kraus operators
    M0 = np.sqrt(1-strength) * np.eye(2) + (np.sqrt(strength) - np.sqrt(1-strength)) * P0
    M1 = np.sqrt(1-strength) * np.eye(2) + (np.sqrt(strength) - np.sqrt(1-strength)) * P1

    # Post-measurement state (outcome 0)
    rho0 = M0 @ rho @ M0.conj().T
    p0 = np.trace(rho0)
    rho0 = rho0 / p0

    # Post-measurement state (outcome 1)  
    rho1 = M1 @ rho @ M1.conj().T
    p1 = np.trace(rho1)
    rho1 = rho1 / p1

    return rho0, p0, rho1, p1

# Start with |+>
rho_plus = np.array([[0.5, 0.5], [0.5, 0.5]])

print("Weak Z measurement of |+>:")
for strength in [0.0, 0.25, 0.5, 0.75, 1.0]:
    rho0, p0, rho1, p1 = weak_z_measurement(rho_plus, strength)
    print(f"  strength={strength:.2f}: P(0)={p0:.3f}, P(1)={p1:.3f}")
    print(f"    After outcome 0: off-diag = {rho0[0,1]:.3f} (coherence)")
    print(f"    After outcome 1: off-diag = {rho1[0,1]:.3f} (coherence)")

4.10 Measurement in Practice: Noise, Imperfections, and Readout Error

Real quantum measurements are not perfect projectors. They suffer from:

  1. Readout error (assignment error): The classical outcome is misidentified. A qubit in $|0\rangle$ might be read as $|1\rangle$ with probability $p(1|0)$, and vice versa. Typical error rates on superconducting qubits are 1-5%.

  2. Measurement crosstalk: Measuring one qubit can affect the measurement of an adjacent qubit. The error matrix is not diagonal.

  3. Projection quality: Real measurements are not infinitely fast. During the measurement time, the state may evolve or decohere.

Readout error mitigation. Given a $2^n \times 2^n$ assignment matrix $M$ where $M_{j,i} = P(\text{read } j | \text{prepared } i)$, the true probability vector $\vec{p}_{\text{true}}$ is related to the observed vector by:

$$\vec{p}_{\text{obs}} = M \vec{p}_{\text{true}}$$

We can invert this: $\vec{p}_{\text{true}} = M^{-1} \vec{p}_{\text{obs}}$, though regularization is needed for noisy data.

# Qiskit: Readout error mitigation
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
from qiskit.result import LocalReadoutMitigator
import numpy as np

# Simulate a simple circuit with readout error
simulator = AerSimulator()
qc = QuantumCircuit(1, 1)
qc.h(0)
qc.measure(0, 0)

# Run without mitigation
result = simulator.run(qc, shots=10000).result()
counts = result.get_counts()
print("Raw counts:", counts)
print("P(0) raw:", counts.get('0', 0) / 10000)

# On real hardware, you would calibrate the assignment matrix
# by preparing |0> and |1> and measuring
# For simulation, the counts should be close to 50/50