Quantum Channels and Open-System Dynamics

Introduction and context

Quantum theory describes an isolated system by a state vector or, more generally, a density operator. Real quantum devices, however, are never perfectly isolated. Qubits interact with control electronics, thermal fields, neighbouring qubits, measurement apparatus and uncontrolled environmental degrees of freedom. These interactions cause noise, loss of coherence and energy relaxation.

Quantum information theory models such physical evolution using quantum channels. A quantum channel is a mathematical transformation that maps an input density operator to a valid output density operator. The channel formalism is sufficiently general to describe ideal gates, noisy communication, storage errors, measurement processes and the effective dynamics of a subsystem embedded in a larger environment.

The central objective of this chapter is to model noise and physical evolution using completely positive trace-preserving maps, Kraus operators and common channels including depolarising, dephasing and amplitude-damping channels.

Learning objectives

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

1. Explain why density operators and quantum channels are required for open-system modelling.
2. Define positive, completely positive, trace-preserving and completely positive trace-preserving maps.
3. Use Kraus operators to calculate the output of a quantum channel.
4. Verify whether a proposed operator-sum representation is trace-preserving.
5. Describe and compare depolarising, dephasing and amplitude-damping channels.
6. Interpret channel parameters physically and geometrically on the Bloch sphere.
7. Model sequential noise processes and identify practical limitations of simple channel models.

1. Quantum states of open systems

A pure qubit state can be written as |ψ⟩ = α|0⟩ + β|1⟩, where |α|² + |β|² = 1. Open-system analysis must also represent classical uncertainty about preparation and entanglement with an environment. The appropriate description is a density operator ρ.

A density operator satisfies three conditions:

• ρ = ρ† (Hermiticity);
• ρ ≥ 0 (positive semidefiniteness);
• Tr(ρ) = 1 (unit trace).

A statistical mixture is represented by ρ = ∑_i p_i |ψ_i⟩⟨ψ_i|, where p_i ≥ 0 and ∑_i p_i = 1. Different ensembles can produce the same density operator; therefore, the density operator, rather than a particular ensemble interpretation, is the physical object used for prediction.

For a single qubit, every density operator has the Bloch representation

Here I is the identity matrix, X, Y and Z are the Pauli operators, r = (r_x,r_y,r_z) is the Bloch vector, and σ = (X,Y,Z). Physical states satisfy |r| ≤ 1. Pure states lie on the surface of the Bloch sphere, while mixed states lie inside it.

An isolated system evolves under a unitary operator U, according to ρ′ = UρU†. This transformation is reversible and preserves the eigenvalues of ρ. Open-system evolution is generally not reversible because information can flow into an environment.

2. Quantum channels

A quantum channel is a linear map E that transforms operators on an input Hilbert space into operators on an output Hilbert space. For a physical channel, the output must be a valid density operator whenever the input is valid.

Positivity and complete positivity

A map E is positive if it maps every positive operator to another positive operator:

ρ ≥ 0 ⇒ E(ρ) ≥ 0.

Positivity alone is insufficient for quantum information. An input system may be entangled with an untouched reference system. A physically valid map must remain positive when it acts on one part of an arbitrarily large composite system. Thus E is completely positive if, for every reference dimension n,

where I_n is the identity map on the reference system. Complete positivity protects the validity of entangled states.

A map is trace-preserving when

for every input operator. A completely positive trace-preserving map, abbreviated CPTP map, is the standard mathematical representation of a deterministic quantum channel.

A trace-non-increasing completely positive map can instead represent one outcome of a measurement or a conditional process. If E_k represents outcome k, then the probability of that outcome is p_k = Tr(E_k(ρ)), and the normalized conditional state is ρ_k = E_k(ρ)/p_k when p_k > 0. The sum over all outcomes is trace-preserving.

Why complete positivity matters

The transpose operation T is positive because it preserves eigenvalues of a single matrix. However, the partial transpose of some entangled density operators is not positive. Therefore, transposition is not completely positive and cannot describe a physical channel acting on an unknown subsystem. This example shows why testing only isolated input states is inadequate.

3. Kraus or operator-sum representation

Every finite-dimensional CPTP map can be represented in Kraus form:

The operators K_k are called Kraus operators or operation elements. They map the input Hilbert space to the output Hilbert space. Trace preservation is equivalent to the completeness relation

This condition is an operator identity, not merely a condition on one selected state. Complete positivity is automatic from the Kraus form because each term K_k ρ K_k† is positive whenever ρ is positive.

Kraus representations are not unique. If {K_k} and {L_j} describe the same channel, they may be related by an isometry; when both sets have the same number of elements, a unitary mixing can connect them. The minimum number of Kraus operators equals the rank of the channel's Choi matrix and is called the channel's Kraus rank.

Physical interpretation

A useful interpretation is that the environment records which interaction branch occurred. The system is transformed by K_k in branch k, but the branch information is not observed. Summing over branches produces the system's reduced state. The operators are not generally probabilities themselves; the probability of branch k is

The corresponding normalized state is K_k ρ K_k†/p_k.

4. Deriving channels from system–environment evolution

Suppose the system begins in state ρ and the environment in state |e₀⟩⟨e₀|. Let a joint unitary U_SE act on system and environment. The final system state is obtained by tracing out the environment:

Choose an orthonormal environment basis [[MATH{|k⟩}]] and define

Then the partial trace gives the Kraus representation E(ρ) = ∑_k K_kρK_k†. The unitarity of U_SE implies ∑_k K_k†K_k = I.

This construction explains why open-system evolution is CPTP: it is unitary evolution on a larger closed system followed by discarding the environment. It also explains why a channel can be irreversible for the subsystem even though the joint evolution is reversible.

5. Important single-qubit channels

5.1 Bit-flip channel

The bit-flip channel applies X with probability p and does nothing with probability 1−p:

A Kraus representation is K₀ = √(1−p)I and K₁ = √p X. The completeness relation is K₀†K₀ + K₁†K₁ = (1−p)I + pI = I.

A bit flip exchanges computational-basis populations. It does not simply mean that every state becomes random: its action depends on the input state and on p.

5.2 Phase-flip and dephasing channels

A phase flip applies Z with probability p:

For

the channel gives

Thus populations remain unchanged while coherences are multiplied by 1−2p. At p = 1/2, all off-diagonal terms vanish. This is complete dephasing in the computational basis.

A commonly used continuous dephasing model is parameterized directly by a coherence factor λ, where 0 ≤ λ ≤ 1:

One Kraus representation is K₀ = √((1+λ)/2)I and K₁ = √((1−λ)/2)Z. Here λ = e^−t/T₂ is often used for exponential pure dephasing, with time t and dephasing time T₂ under a Markovian model. The exact relation between measured T₂ and other relaxation times depends on the physical noise model.

5.3 Depolarising channel

The qubit depolarising channel replaces the input with the maximally mixed state with probability p and otherwise leaves it unchanged:

For the conventional range 0 ≤ p ≤ 1, the Bloch vector changes as r′ = (1−p)r. A Kraus representation is

The completeness relation holds because the squared coefficients sum to one. Using the Pauli identities, this representation yields E_dep(ρ) = (1−p)ρ + pI/2. At p = 1, every input becomes I/2.

Parameter conventions vary in textbooks. Some define the depolarising channel by assigning probability q to each of the three Pauli errors, giving E(ρ) = (1−3q)ρ + q(XρX + YρY + ZρZ). In that convention, the replacement parameter above is p = 4q. Parameters must therefore be identified before comparing reported error rates.

5.4 Amplitude-damping channel

Amplitude damping models energy relaxation, such as spontaneous emission from |1⟩ to |0⟩. With damping probability γ,

The completeness relation is K₀†K₀ + K₁†K₁ = I. For a general qubit density matrix,

The excited-state population decreases by a factor 1−γ, and the ground-state population increases correspondingly. Unlike a unital channel, amplitude damping does not preserve the maximally mixed state. It drives the system toward the ground state.

In a simple Markovian relaxation model, γ(t) = 1−e^−t/T₁, where T₁ is the energy-relaxation time. The coherence factor due solely to this relaxation is e^−t/(2T₁). Additional pure dephasing may reduce coherence further. In many physical systems, the measured transverse decay satisfies approximately

where T_φ is the pure-dephasing time, assuming independent exponential Markovian contributions.

Comparison of common channels

6. Worked example: applying amplitude damping

Consider

Let γ = 0.36, so 1−γ = 0.64 and √(1−γ) = 0.8. Substitution gives

The trace is 0.68 + 0.32 = 1, as required. The excited-state population has decreased from 0.5 to 0.32, while the ground-state population has increased to 0.68. The off-diagonal coherence has decreased from 0.5 to 0.40.

The output is mixed. Its determinant is 0.68×0.32−0.40² = 0.0576, which is positive, confirming that it is not a pure state.

7. Bloch-sphere interpretation

Quantum channels transform the Bloch vector by an affine transformation

where A is a real 3 × 3 matrix and c is a real translation vector. For unital channels, c = 0; the maximally mixed state remains fixed. Depolarising noise contracts the sphere equally in all directions. Dephasing contracts the transverse x and y components while leaving z unchanged. Amplitude damping both contracts and translates the Bloch sphere toward the ground-state pole.

This geometric view is valuable for diagnosing hardware. An isotropic contraction suggests depolarising-like noise, whereas directional contraction suggests dephasing or relaxation. Real devices commonly exhibit combinations of these effects.

8. Sequential channels and composition

If a state passes through E₁ followed by E₂, the total channel is

If E₁ has Kraus operators {A_i} and E₂ has Kraus operators {B_j}, then a Kraus set for the composition is {B_jA_i}. The order matters: in general, E₂ ∘ E₁ ≠ E₁ ∘ E₂.

For example, dephasing before a computational-basis measurement affects interference but does not change computational populations. Amplitude damping before a gate changes both populations and later interference behaviour. Consequently, a circuit-level error estimate must account for where the channel occurs, not merely for its average probability.

A channel can also be extended with an identity operation on another subsystem. If a noise process acts independently on two qubits, the joint channel is often approximated by E_A ⊗ E_B. This assumption can fail when the qubits share correlated environmental fluctuations.

9. Choi representation and channel verification

For finite-dimensional systems, complete positivity can be tested using the Choi matrix. Given a channel E on a d-dimensional system, define

The map is completely positive if and only if J(E) ≥ 0. It is trace-preserving if and only if

The Choi matrix is particularly useful in process tomography and numerical channel reconstruction. Experimental data are used to estimate a process matrix, which must then be constrained to be positive and trace-preserving. Without these constraints, statistical noise can produce an unphysical reconstructed map.

10. Illustrative case: diagnosing a superconducting-qubit memory

Background. A laboratory stores a qubit for a controlled interval and measures its state afterward. The engineering team observes that computational-basis populations decay slowly, while interference fringes disappear more rapidly.

Problem. The team must decide whether the memory is dominated by energy relaxation, pure dephasing or an approximately isotropic error process.

Analysis. They prepare |0⟩, |1⟩, |+⟩ and |+_y⟩, and repeat the experiment for several storage times. Preparations in the computational basis estimate population relaxation and therefore T₁. Superposition states probe transverse coherence and estimate T₂. If the excited population follows P₁(t) = e^−t/T₁ and the transverse coherence follows e^−t/T₂, then the approximate relation 1/T₂ = 1/(2T₁) + 1/T_φ gives an estimate of pure dephasing.

Approach. The team fits an amplitude-damping channel to the population data and adds a dephasing channel to account for the remaining coherence loss. They then perform process tomography at selected storage times and enforce CPTP constraints during reconstruction.

Outcome. The fitted model shows that relaxation accounts for part, but not all, of the loss of coherence. The resulting noise model predicts memory performance more accurately than a depolarising approximation, especially for algorithms whose logical information is encoded in energy populations.

Learning points. First, different preparations reveal different channel components. Second, a single average error rate can conceal directional noise. Third, a model should be validated against data not used in fitting. Finally, the inferred channel is an effective model over a specified time range; it should not automatically be treated as a universal microscopic law.

11. Practical modelling, limitations and responsible use

Markovian versus non-Markovian dynamics

A Markovian model assumes that the future evolution depends only on the current state and that environmental memory can be neglected. It often leads to exponential decay and a composition law with time-homogeneous channels. Real environments can retain information and feed it back into the system, producing non-exponential decay, correlations between time intervals and even temporary increases in distinguishability.

Correlated and leakage errors

Independent single-qubit channels are convenient but may be inaccurate. Shared control lines, common fluctuating fields and crosstalk create correlated errors. Leakage is also important: a physical device may leave the intended two-level computational subspace. A two-dimensional CPTP model cannot fully represent leakage unless the leakage states are included or an effective loss model is explicitly justified.

Parameter conventions and identifiability

Different channels can produce similar results for a restricted set of input states. Reliable identification therefore requires informationally complete preparations and measurements, repeated trials, uncertainty estimates and independent validation. Report the channel convention, parameter range, basis choice, time interval and whether the parameter is a probability, a decay factor or a rate.

Professional and governance considerations

Noise models influence hardware comparisons, fault-tolerance estimates and resource planning. Researchers should not report a fitted channel without uncertainty or hide data-selection criteria. In safety-critical or high-value quantum applications, model assumptions should be documented, calibration drift should be monitored and conclusions should be reproducible. Privacy and security also matter: process data can reveal device characteristics, calibration weaknesses or side-channel information.

12. Best practices

1. Start with the simplest CPTP model consistent with the evidence.
2. Verify ∑_k K_k†K_k = I analytically or numerically.
3. State the basis in which dephasing or relaxation is defined.
4. Distinguish a channel probability, a coherence factor and a physical decay time.
5. Test models on states and circuits not used for calibration.
6. Include correlated noise and leakage when the experiment indicates them.
7. Use confidence intervals or posterior distributions for estimated parameters.
8. Record parameter conventions so results can be compared across systems.
9. Avoid interpreting an effective channel as a complete microscopic explanation.

Chapter Summary

Quantum channels provide the framework for describing non-unitary evolution of quantum systems. A physical deterministic channel is a completely positive trace-preserving map: complete positivity preserves validity when the system is entangled with a reference, while trace preservation guarantees total probability remains one. Every finite-dimensional CPTP map has a Kraus representation E(ρ) = ∑_k K_kρK_k†, with completeness condition ∑_k K_k†K_k = I.

System–environment unitary evolution followed by an environmental partial trace naturally produces this form. Dephasing preserves populations but suppresses coherences; depolarising noise contracts all Bloch-vector directions toward the maximally mixed state; amplitude damping models energy relaxation and translates states toward the ground state. Channels compose sequentially, may be analysed geometrically through affine Bloch transformations and can be verified using Choi matrices. Practical noise modelling requires attention to parameter conventions, correlated errors, leakage, non-Markovian effects, uncertainty and responsible reporting.

Key Terms

Density operator: A positive, Hermitian, unit-trace operator representing a quantum state.

Open system: A quantum system that interacts with degrees of freedom outside the system of interest.

Quantum channel: A physical transformation of quantum states, usually represented by a CPTP map.

Positive map: A map that sends positive operators to positive operators.

Completely positive map: A map that remains positive when extended by an identity map on any reference system.

Trace-preserving map: A map that preserves the trace of every operator.

CPTP map: A completely positive trace-preserving map.

Kraus operator: An operator in an operator-sum representation of a quantum channel.

Operator-sum representation: The expression E(ρ) = ∑_k K_kρK_k†.

Kraus rank: The minimum number of Kraus operators required for a channel.

Dephasing: Loss of phase coherence without necessarily changing computational-basis populations.

Depolarising channel: A channel that contracts the Bloch vector isotropically toward the maximally mixed state.

Amplitude damping: A channel describing energy relaxation, commonly from |1⟩ to |0⟩.

Unital channel: A channel satisfying E(I) = I.

Bloch vector: The three-dimensional vector representing a single-qubit density operator.

Choi matrix: A matrix whose positivity characterises complete positivity of a finite-dimensional map.

Markovian dynamics: Evolution modelled without relevant environmental memory.

Leakage: Population leaving the intended computational Hilbert space.

Process tomography: Experimental reconstruction of a quantum channel from input-output data.

Review Questions

Short Answer Questions

1. What two conditions distinguish a CPTP map from a general linear map?
2. State the Kraus completeness relation for a trace-preserving channel.
3. What physical information is lost during dephasing?
4. Why is amplitude damping non-unital?
5. What does the Choi matrix test?

Long Answer Questions

1. Explain how a system–environment unitary interaction leads to a Kraus representation.
2. Compare bit-flip, dephasing, depolarising and amplitude-damping channels.
3. Explain why positivity alone is insufficient for quantum channels.
4. Derive the output of the amplitude-damping channel for a general qubit density matrix.
5. Discuss the assumptions and limitations of Markovian noise modelling.

Analytical / Application Questions

1. A qubit undergoes dephasing with coherence factor λ = 0.7. Determine the output density matrix for an input state with diagonal entries 0.6 and 0.4 and off-diagonal entry 0.3.
2. A memory experiment estimates T₁ = 80 μs and T₂ = 50 μs. Assuming independent exponential Markovian relaxation and pure dephasing, calculate T_φ using 1/T₂ = 1/(2T₁) + 1/T_φ.
3. Design an experimental sequence that distinguishes isotropic depolarising noise from dephasing and amplitude damping. Identify the input states and measurements required.

Multiple Choice Questions with Answers

Question 1

Which condition ensures that a quantum map preserves total probability?

A. Positivity

B. Complete positivity

C. Trace preservation

D. Unitality

Correct Answer: C. Trace preservation

Question 2

For Kraus operators to represent a trace-preserving channel, which relation must hold?

A. ∑_k K_kK_k† = 0

B. ∑_k K_k†K_k = I

C. K_k = K_k† for every k

D. ∑_k K_k = I

Correct Answer: B. ∑_k K_k†K_k = I

Question 3

What is the main effect of a dephasing channel in the computational basis?

A. It exchanges |0⟩ and |1⟩ populations

B. It increases all coherences

C. It suppresses off-diagonal terms while preserving populations

D. It always maps the state to |0⟩

Correct Answer: C. It suppresses off-diagonal terms while preserving populations

Question 4

Which channel models spontaneous decay from |1⟩ to |0⟩?

A. Amplitude damping

B. Pure bit flip

C. Depolarising

D. Phase flip

Correct Answer: A. Amplitude damping

Question 5

For the depolarising channel E(ρ) = (1−p)ρ + pI/2, what is the output when p = 1?

A. The input state ρ

B. The state |0⟩⟨0|

C. The maximally mixed state I/2

D. A state with doubled coherence

Correct Answer: C. The maximally mixed state I/2

Question 6

A map is completely positive if it remains positive when tensored with:

A. A zero operator only

B. An identity map on an arbitrary reference system

C. A transpose map only

D. A non-unitary measurement only

Correct Answer: B. An identity map on an arbitrary reference system

Question 7

If amplitude damping has γ = 0.25 and the initial excited-state population is 0.8, what is the final excited-state population?

A. 0.20

B. 0.55

C. 0.60

D. 1.05

Correct Answer: C. 0.60

Question 8

Which channel is generally non-unital?

A. Amplitude damping

B. Depolarising noise

C. A unitary channel

D. The identity channel

Correct Answer: A. Amplitude damping

Question 9

In a sequential process where E₁ acts first and E₂ acts second, the total channel is:

A. E₁ + E₂

B. E₁ − E₂

C. E₁ ∘ E₂ with E₂ acting first

D. E₂ ∘ E₁

Correct Answer: D. E₂ ∘ E₁

Question 10

For a qubit undergoing pure dephasing with λ = 0.5, an off-diagonal density-matrix element initially equal to 0.4 becomes:

A. 0.2

B. 0.4

C. 0.8

D. 0

Correct Answer: A. 0.2