Mixed States, Density Operators and Open Quantum Systems

Introduction

Quantum mechanics is often introduced through pure states, represented by state vectors such as |ψ⟩. This description is powerful when a system is prepared in a known state and evolves without uncontrolled interaction. Computer scientists, however, frequently work with incomplete information, statistical ensembles, subsystems of larger systems, noise, measurement, and environmental interaction. State vectors alone cannot describe these situations adequately.

The density operator, or density matrix, provides the general language for quantum states. It represents both classical uncertainty about preparation and quantum uncertainty arising from entanglement with an unobserved subsystem. Density operators also provide the natural starting point for partial traces, quantum channels, decoherence, dissipation, quantum error correction, and realistic quantum-device modelling.

This chapter develops the formalism at diploma level, while connecting it to quantum computing practice. The emphasis is on interpreting the mathematics correctly: a mixed state is not merely a poorly known pure state, and environmental noise is not generally equivalent to a single unitary operation on the system.

Learning objectives

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

• distinguish pure states, proper mixtures, and improper mixtures;
• construct and interpret density operators;
• test whether an operator is a valid quantum state;
• calculate expectation values, purity, and von Neumann entropy;
• obtain subsystem states using the partial trace;
• describe open-system evolution using quantum channels and Kraus operators;
• explain decoherence, dissipation, and measurement as physical processes;
• model common one-qubit noise channels;
• identify limitations and practical implications of open-system models.

1. Pure states and statistical ensembles

A pure quantum state is represented by a normalized vector |ψ⟩ in a Hilbert space. Normalization requires ⟨ψ|ψ⟩ = 1. A vector and any nonzero global phase multiple represent the same physical state: |ψ⟩ and e^iφ|ψ⟩ make identical predictions.

A pure state gives the probability of an outcome associated with projector Π through the Born rule:

A different situation occurs when a preparation procedure produces |ψ_i⟩ with classical probability p_i, where p_i ≥ 0 and ∑_i p_i = 1. The system is then described by the density operator

The ensemble decomposition is not unique. Different preparation procedures can produce the same operator. Consequently, the density operator—not a particular list of states and probabilities—is the operational description relevant to all measurements.

For example, an equal mixture of |0⟩ and |1⟩ has

This state is the maximally mixed state of one qubit. It is not the same as |+⟩ = (|0⟩ + |1⟩)/√2, even though both give equal probabilities when measured in the computational basis. The superposition |+⟩ also produces interference in another basis; the mixture does not.

Proper and improper mixtures

A proper mixture represents classical uncertainty about which preparation was made. An improper mixture arises when a system is entangled with another system and the second system is ignored. Mathematically both are density operators, and measurements confined to the system cannot distinguish them. Physically, their origins differ, which matters when analysing correlations and possible purification.

2. Density operators and their properties

For a finite-dimensional system, a density operator is a linear operator ρ satisfying three conditions:

1. Hermiticity: ρ = ρ†.
2. Positive semidefiniteness: ⟨φ|ρ|φ⟩ ≥ 0 for every vector |φ⟩.
3. Unit trace: Tr(ρ) = 1.

Hermiticity guarantees real measurement probabilities and real expectation values for observables. Positivity guarantees that probabilities cannot be negative. Unit trace guarantees that all probabilities sum to one.

In a chosen orthonormal basis, a density operator is a matrix. For a qubit,

where ρ10 = ρ01*, ρ00 ≥ 0, ρ11 ≥ 0, and ρ00 + ρ11 = 1. Positivity additionally requires det(ρ) ≥ 0.

The probability of an outcome represented by projector Π_a is

For an observable A, its expectation value is

These expressions include pure states as a special case because ρ = |ψ⟩⟨ψ| gives Tr(ρA) = ⟨ψ|A|ψ⟩.

Spectral decomposition

Every density operator has a spectral decomposition

where the eigenvectors |k⟩ are orthonormal and the eigenvalues satisfy λ_k ≥ 0 and ∑_k λ_k = 1. The eigenvalues can therefore be interpreted as probabilities in an orthogonal ensemble.

A state is pure exactly when ρ² = ρ, equivalently when Tr(ρ²) = 1, or when one eigenvalue is 1 and all others are zero. For a d-dimensional system, the maximally mixed state is I/d, with purity Tr((I/d)²) = 1/d. Thus purity satisfies

The lower bound applies to a d-dimensional state; the upper bound applies to all states.

3. The Bloch-vector representation

Every one-qubit density operator can be written as

where X, Y, and Z are the Pauli matrices, r = (r_x,r_y,r_z) is a real Bloch vector, and σ = (X,Y,Z). Physicality requires |r| ≤ 1. Pure states lie on the unit sphere, while mixed states lie inside it. The completely mixed state is at the origin.

The components have direct measurement meaning:

The purity becomes

Thus environmental noise generally shortens the Bloch vector. Unitary evolution rotates it without changing its length.

4. Entropy and information

The von Neumann entropy quantifies the mixedness of a state:

The entropy is measured in bits because the logarithm is base 2. The convention 0 log₂0 = 0 is used. A pure state has entropy zero, and the maximally mixed state of dimension d has entropy log₂d.

Entropy is not identical to ignorance in every interpretation. In quantum information it is an operational quantity associated with the spectrum of ρ. For a composite system, the entropy of a subsystem can be nonzero even when the total state is pure, because entanglement transfers information into correlations.

5. Composite systems and the partial trace

For systems A and B with state spaces of dimensions d_A and d_B, the composite space is the tensor product H_A ⊗ H_B. A joint density operator ρ_AB describes the complete state. If only A is accessible, its reduced state is obtained by tracing out B:

In an orthonormal basis [[MATH{|j⟩_B}]],

Similarly, ρ_B = Tr_A(ρ_AB). The defining operational property is

for every observable A acting on A.

Worked example: an entangled pair

Consider the Bell state

Its joint density operator contains diagonal terms |00⟩⟨00| and |11⟩⟨11| together with coherence terms |00⟩⟨11| and |11⟩⟨00|. Tracing out the second qubit removes the cross-system labels and gives

Therefore each qubit individually is maximally mixed, although the pair is in a pure state. The missing information is not random preparation noise; it is encoded in correlations between the qubits.

6. Quantum operations and channels

A physical process acting on density operators must map valid states to valid states. The general mathematical model is a completely positive trace-preserving map (CPTP map), also called a quantum channel:

Complete positivity means that 𝓔 ⊗ I_R remains positive for every reference system R. This condition is essential because a system may be entangled with another system. Positivity alone is insufficient.

Trace preservation means Tr(𝓔(ρ)) = Tr(ρ). A channel has a Kraus representation

with the completeness condition

The operators E_k are Kraus operators. A channel may have more than one valid Kraus representation; the physical map is the invariant object.

A unitary operation is the special case with one Kraus operator E₁ = U, where U†U = I:

Unitary evolution preserves eigenvalues, purity, and von Neumann entropy. General channels need not preserve them.

Measurement as a quantum operation

A projective measurement with projectors Π_m satisfies Π_mΠ_n = δ_mnΠ_m and ∑_m Π_m = I. The probability of outcome m is

If the outcome is recorded but the result is ignored, the post-measurement state is

This is a dephasing operation in the measurement basis. If the outcome is known, the conditional state is ρ_m = Π_mρΠ_m/p(m), provided p(m) > 0.

7. Open quantum systems

An open quantum system interacts with an environment that is not fully controlled or observed. Let S denote the system and E the environment. If the initial state is a product state ρ_S ⊗ ρ_E and the joint evolution is unitary U_SE, then the system evolves as

This expression explains why non-unitary system dynamics can arise from globally unitary physics. The environment carries away information about the system, and tracing it out produces a channel on S.

The product-state assumption is convenient but not universal. If S and E are initially correlated, the reduced evolution may fail to be representable by a single CPTP map acting on every possible system input. This is an important limitation in realistic memory-bearing environments.

Decoherence

Decoherence is the suppression of phase coherence in a preferred basis caused by environmental correlations. Suppose

evolves approximately to

The reduced system state is

If the environmental states become orthogonal, ⟨e₁|e₀⟩ = 0, the off-diagonal terms vanish. Populations remain, but interference is lost. Decoherence does not necessarily transfer energy; it primarily destroys phase information.

Dissipation and relaxation

Dissipation involves energy exchange with the environment. A qubit may relax from |1⟩ to |0⟩, releasing energy, or experience thermal excitation. Decoherence and dissipation can occur together, but they are conceptually distinct: a system may dephase with little energy change, or relax while also losing phase coherence.

8. Important single-qubit noise channels

The following channels are idealized models. Their parameters depend on the physical device and the time interval represented.

Bit-flip channel

Here p is the probability of applying the Pauli-X error, with 0 ≤ p ≤ 1.

Phase-flip channel

This changes relative phase information without exchanging computational-basis population. For a density matrix, the off-diagonal terms are multiplied by 1−2p.

Depolarizing channel

One common convention is

for a qubit, where 0 ≤ p ≤ 1. It contracts the Bloch vector by 1−p. Other literature uses a different parameterization involving probabilities for X, Y, and Z errors; parameters must not be compared without checking the convention.

Amplitude damping

Amplitude damping models energy relaxation toward |0⟩. With damping probability γ,

and

The completeness relation is E₀†E₀ + E₁†E₁ = I. Unlike a unital channel, amplitude damping maps the maximally mixed state toward the ground state.

Phase damping

Phase damping suppresses off-diagonal entries while leaving computational-basis populations unchanged. It is useful as a phenomenological model of dephasing, although real devices may exhibit non-Markovian or frequency-dependent noise that this simple model does not capture.

9. Worked calculation: amplitude damping

Let

and let γ = 0.20. Using the amplitude-damping operators above, the output matrix is

Therefore,

• ρ'₀₀ = 0.25 + 0.20(0.75) = 0.40;
• ρ'₁₁ = 0.80(0.75) = 0.60;
• ρ'₀₁ = √0.80(0.25) ≈ 0.2236.

Thus

The trace remains 1. The excited-state population decreases from 0.75 to 0.60, while the lost population appears in the ground state. The coherence is also reduced by the factor √(1−γ).

10. Case study: modelling noise in a small quantum algorithm

Background

This illustrative case concerns a two-qubit educational quantum processor used to demonstrate a Bell-state circuit. The circuit applies a Hadamard gate to qubit A followed by a controlled-NOT gate with A as control and B as target. Ideally, the output is |Φ⁺⟩.

Problem

Experimental measurements show that each qubit individually appears close to maximally mixed, while Bell-basis measurements reveal fewer ideal correlations than expected. Students must decide whether the result is consistent with classical preparation uncertainty, decoherence, or an incorrect gate operation.

Analysis

The ideal joint state is pure, but each reduced state is I/2. Therefore single-qubit statistics alone cannot confirm successful entanglement. A noise model is required. Suppose each qubit undergoes a phase-flip channel after the circuit. Phase errors reduce the coherence between |00⟩ and |11⟩, weakening Bell correlations while leaving computational-basis populations approximately unchanged.

A useful diagnostic sequence is:

1. perform calibration measurements to estimate readout bias;
2. compare computational-basis populations with Bell-basis correlations;
3. reconstruct an estimated density matrix using state tomography;
4. test whether the estimate is positive semidefinite and trace one;
5. compare candidate channels, such as phase-flip, depolarizing, and amplitude-damping models;
6. repeat at different circuit durations to distinguish gate error from time-dependent decoherence.

Approach and outcome

If populations remain stable but off-diagonal estimates decay with time, phase damping is a plausible first model. If the excited-state population decreases, amplitude damping must also be included. In practice, a fitted model may be represented as a composition such as 𝓔_total = 𝓔_readout ∘ 𝓔_damping ∘ 𝓔_dephasing, although the order and validity of this composition must be experimentally justified.

The outcome is not simply a single “error percentage.” The density matrix identifies which information is lost, while channel parameters support mitigation, error correction, and hardware comparison.

Learning points

• reduced states can hide entanglement;
• computational-basis agreement does not prove coherent operation;
• tomography and channel characterization require carefully designed measurements;
• a noise model is an approximation whose validity must be tested against data;
• error mitigation cannot restore information that has irreversibly leaked without additional assumptions.

11. Practical computation and implementation

In software, a density matrix for a d-dimensional system requires d² complex entries, compared with d complex amplitudes for a state vector. For n qubits, d = 2ⁿ, so direct density-matrix simulation scales as 4ⁿ matrix entries. This exponential cost limits simulation size.

Practical strategies include:

• quantum trajectories or Monte Carlo wave functions for some open-system models;
• tensor-network methods when entanglement and interaction structure are limited;
• stabilizer methods for restricted Clifford circuits and selected noise models;
• process tomography for small devices;
• randomized benchmarking for operational estimates of gate quality;
• noise representations adapted to sparse or structured channels.

Tomography is resource-intensive because many measurement settings are required and statistical noise can produce an unphysical raw estimate. Reconstruction methods commonly impose Hermiticity, positivity, and unit trace. These constraints are not optional stylistic choices; they define the state space.

12. Limitations and contemporary developments

The Markovian channel model assumes that the environment does not retain relevant memory of earlier system states. This approximation leads to time-local master equations and simple repeated-channel models, but it can fail in structured environments, strong coupling, low temperatures, or slowly varying noise.

A widely used continuous-time model is the Lindblad master equation:

Here H is the system Hamiltonian, L_k are Lindblad operators, [H,ρ] = Hρ−ρH is the commutator, {A,B} = AB+BA is the anticommutator, and ℏ is the reduced Planck constant. The equation assumes a Markovian, time-homogeneous approximation in its standard form. The dissipative term preserves trace and complete positivity under the usual Lindblad conditions.

Contemporary quantum engineering increasingly combines density-operator models with machine learning, Bayesian estimation, and real-time calibration. Such tools can help infer noise parameters, but they do not remove the need for physical constraints, uncertainty quantification, and independent validation. A flexible statistical fit can reproduce data while still lacking predictive physical meaning.

13. Best practices for quantum-system modelling

1. State the Hilbert space, basis, and subsystem ordering explicitly.
2. Check Hermiticity, positivity, and trace before interpreting a density matrix.
3. Declare channel parameter conventions; names such as “depolarizing probability” are not universal.
4. Distinguish conditional post-measurement states from non-selective measurement states.
5. Separate preparation uncertainty, gate noise, measurement noise, and environmental decoherence.
6. Verify that a proposed map is completely positive, not merely trace preserving.
7. Report assumptions about initial system–environment correlations.
8. Compare model predictions with measurements outside the data used for fitting.
9. Preserve uncertainty estimates in tomography and parameter estimation.
10. Avoid claiming entanglement from reduced-state measurements alone.

Ethical and governance considerations

Quantum-device characterization involves experimental data, proprietary hardware information, and sometimes claims about computational advantage. Researchers and engineers should report calibration conditions, discarded data, model assumptions, and statistical uncertainty honestly. Overstating fidelity or hiding readout errors can mislead users and distort comparisons between platforms.

When quantum systems are used in security, optimization, or scientific decision-making, noise models should be treated as part of the evidence chain. Reproducible code, documented parameter conventions, and independent validation are especially important because different channel models can produce similar results on limited experiments.

Chapter Summary

Density operators generalize state vectors to statistical ensembles, entangled subsystems, measurements, and noisy processes. A valid density operator is Hermitian, positive semidefinite, and trace one. Pure states satisfy ρ² = ρ and have unit purity; mixed states have lower purity and may possess nonzero von Neumann entropy.

The partial trace produces the state accessible to a subsystem and explains why a subsystem of a pure entangled state can be mixed. Quantum channels are completely positive, trace-preserving maps, represented conveniently by Kraus operators. Open-system evolution results from unitary system–environment dynamics followed by tracing out the environment. Decoherence suppresses phase coherence, while dissipation involves energy exchange.

Bit-flip, phase-flip, depolarizing, amplitude-damping, and phase-damping channels are useful models, but their assumptions and parameter conventions must be stated. Realistic quantum computing requires density-matrix reasoning for noise characterization, tomography, error mitigation, and hardware evaluation.

Key Terms

Density operator: An operator representing a quantum state, including pure and mixed states.

Pure state: A state representable by one normalized state vector.

Mixed state: A state represented by a non-idempotent density operator or a statistical ensemble.

Proper mixture: A mixture arising from classical uncertainty about preparation.

Improper mixture: A reduced state produced by ignoring part of an entangled system.

Hermitian operator: An operator equal to its conjugate transpose.

Positive semidefinite: An operator whose expectation value is nonnegative for every vector.

Purity: The quantity Tr(ρ²) measuring departure from maximal mixedness.

Von Neumann entropy: The quantity S(ρ) = −Tr(ρ log₂ρ) measuring quantum-state entropy.

Bloch vector: A three-dimensional vector representing a one-qubit density operator.

Partial trace: The operation that obtains a subsystem state by disregarding another subsystem.

Quantum channel: A completely positive, trace-preserving map on density operators.

Kraus operator: An operator in a representation 𝓔(ρ)=∑_kE_kρE_k†.

Decoherence: Loss of phase coherence through environmental interaction.

Dissipation: Energy or excitation exchange with an environment.

Amplitude damping: A channel modelling relaxation toward a ground state.

Dephasing: Loss of relative phase information with populations approximately preserved.

Lindblad equation: A standard Markovian master equation for continuous open-system dynamics.

Quantum tomography: Experimental reconstruction of a quantum state or process.

Review Questions

Short Answer Questions

1. What three mathematical properties must every density operator satisfy?
2. Why can two different ensemble decompositions represent the same mixed state?
3. What physical operation is represented by a partial trace?
4. State the difference between decoherence and dissipation.
5. What condition must Kraus operators satisfy for a channel to preserve trace?

Long Answer Questions

1. Explain why the equal mixture of |0⟩ and |1⟩ is physically different from the pure state |+⟩.
2. Derive the reduced state of one qubit from the Bell state |Φ⁺⟩.
3. Describe how a unitary system–environment interaction can produce non-unitary dynamics for the system.
4. Compare phase-flip, depolarizing, and amplitude-damping channels, including the information each tends to destroy.
5. Explain the role of complete positivity and why positivity alone is not sufficient for a quantum channel.

Analytical / Application Questions

1. A one-qubit state has Bloch vector r=(0.6,0.0,0.8). Determine whether it is pure or mixed and calculate its purity.
2. Design an experiment to distinguish phase damping from amplitude damping using measurements in at least two bases. State the expected signatures.
3. A tomography experiment produces a Hermitian matrix with unit trace but one negative eigenvalue. Analyse what this means and propose a physically valid reconstruction strategy.

Multiple Choice Questions with Answers

Question 1

Which condition is required of every density operator?

A. It must be unitary

B. It must have trace zero

C. It must be positive semidefinite and have unit trace

D. It must be diagonal in every basis

Correct Answer: C. It must be positive semidefinite and have unit trace

Question 2

What is the density operator for an equal mixture of |0⟩ and |1⟩?

A. |+⟩⟨+|

B. I/2

C. |0⟩⟨1|

D. Z/2

Correct Answer: B. I/2

Question 3

What does the partial trace over subsystem B produce from ρ_AB?

A. A state of subsystem A

B. A unitary acting on B

C. The determinant of ρ_AB

D. A measurement outcome with certainty

Correct Answer: A. A state of subsystem A

Question 4

For a pure state density operator ρ, which equality holds?

A. ρ² = 0

B. Tr(ρ²) = 0

C. ρ² = ρ

D. Tr(ρ) = 0

Correct Answer: C. ρ² = ρ

Question 5

Which expression gives the probability of a projector measurement Π on state ρ?

A. Tr(ρΠ)

B. det(ρΠ)

C. Tr(ρ+Π)

D. ρΠρ

Correct Answer: A. Tr(ρΠ)

Question 6

What is the Kraus completeness condition for a trace-preserving channel?

A. ∑k Ek = I

B. ∑k EkEk† = 0

C. ∑k Ek†Ek = I

D. Ek†Ek = ρ

Correct Answer: C. ∑k Ek†Ek = I

Question 7

Which process primarily suppresses phase coherence without necessarily exchanging energy?

A. Decoherence or dephasing

B. Normalization

C. Unitary basis change

D. State preparation only

Correct Answer: A. Decoherence or dephasing

Question 8

For the qubit depolarizing convention 𝓔(ρ)=(1−p)ρ+pI/2, what happens to the Bloch vector?

A. It is rotated but keeps its length

B. It is multiplied by 1−p

C. It is multiplied by 1+p

D. It becomes a complex vector

Correct Answer: B. It is multiplied by 1−p

Question 9

In amplitude damping, where does population typically move?

A. From the ground state to an inaccessible basis

B. From the excited state toward the ground state

C. Only between phase components

D. It is always converted into entanglement

Correct Answer: B. From the excited state toward the ground state

Question 10

A qubit has Bloch vector length 0. Which state does this represent?

A. A maximally mixed state

B. A pure computational-basis state

C. An invalid state

D. A unitary operator

Correct Answer: A. A maximally mixed state