Noise and Open Quantum Systems

Introduction and context

An ideal quantum circuit assumes that qubits are perfectly isolated and that every gate, measurement, and reset is exact. Real devices cannot satisfy these assumptions. Qubits interact with electromagnetic fields, control electronics, neighbouring qubits, residual thermal energy, fabrication defects, and the measurement apparatus. These interactions alter quantum states and reduce the reliability of computation.

The study of noise is therefore central to quantum programming and simulation. A useful noise model should do more than add random errors to a circuit: it should represent how physical processes transform density matrices, predict observable changes, support simulation, and connect with calibration data. This chapter develops that framework using quantum channels, Kraus operators, practical noise models, and device-level reasoning.

Learning objectives

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

1. Explain why open quantum systems require density matrices and quantum channels.
2. Distinguish bit-flip, phase-flip, depolarizing, relaxation, dephasing, readout, and gate errors.
3. Verify whether a set of Kraus operators represents a valid trace-preserving channel.
4. Simulate the qualitative and quantitative effects of noise on simple circuits.
5. Interpret calibration quantities such as T₁, T₂, gate error, and readout assignment error.
6. Explain how connectivity, crosstalk, and hardware topology affect noisy-circuit results.
7. Choose an appropriate noise model and recognize its limitations.

1. Closed and open quantum systems

A closed quantum system evolves by a unitary operator U, so a state vector changes according to |ψ'⟩ = U|ψ⟩. A density operator provides a more general description. For a pure state, ρ = |ψ⟩⟨ψ|; for a statistical mixture, ρ = ∑_i p_i |ψ_i⟩⟨ψ_i|, where p_i ≥ 0 and ∑_i p_i = 1.

A density matrix must satisfy three conditions:

• Hermiticity: ρ = ρ†.
• Positive semidefiniteness: measurement probabilities cannot be negative.
• Unit trace: Tr(ρ) = 1.

An open quantum system is one that interacts with an environment. If the system and environment together evolve unitarily but the environment is not observed, the system generally undergoes a non-unitary transformation. The most general physically valid transformation is a completely positive trace-preserving map, commonly called a quantum channel:

Here E_k are Kraus operators. Trace preservation requires

Complete positivity ensures that the transformation remains physically valid even when the system is entangled with another system. A channel is unital if 𝓔(I) = I; not all physical channels are unital. Relaxation, for example, drives a qubit toward its ground state and is non-unital.

The Kraus representation is not unique. Different sets of Kraus operators can represent the same channel. This is useful computationally: a simulator may use Kraus trajectories, density-matrix evolution, or a stochastic error approximation while representing the same physical process.

2. Important categories of quantum noise

2.1 Bit-flip and phase-flip channels

A bit flip changes computational basis states, while a phase flip changes the relative phase. With probability p, the bit-flip channel is

where X = [[0,1],[1,0]]. Its Kraus operators are √(1−p)I and √p X.

The phase-flip channel is

where Z = [[1,0],[0,−1]]. A phase flip does not change computational-basis populations, but it changes interference and therefore affects algorithms that depend on relative phase.

2.2 Depolarizing noise

A single-qubit depolarizing channel replaces a state with the maximally mixed state with probability p:

An equivalent Pauli form is

provided the same parameter convention is used consistently. The first expression defines p as the replacement probability; the second uses Pauli-error probabilities p/3. Confusing these conventions is a common simulation error.

Depolarization shrinks the Bloch vector equally in all directions. It is convenient for benchmarking but often oversimplifies real hardware, where errors are biased and correlated.

2.3 Amplitude damping and relaxation

Amplitude damping represents energy loss, such as decay from |1⟩ to |0⟩. Let γ be the probability of decay during a specified time interval. Kraus operators are

and

For an initially excited qubit, the excited-state population after time t is commonly modelled as

where T₁ is the energy-relaxation time. Thus, for a qubit initially in |1⟩, γ = 1 − e^(−t/T₁). This exponential model assumes a constant relaxation rate and a stationary environment.

Amplitude damping changes populations and also reduces coherence. It is non-unital because the maximally mixed state is not generally preserved.

2.4 Dephasing

Dephasing destroys phase coherence without necessarily changing computational-basis populations. A simple phase-damping model can be written as

where T_φ is the pure-dephasing time and ρ₀1 is an off-diagonal density-matrix element. In many devices, relaxation and pure dephasing combine according to

Here T₂ is the measured transverse-coherence time. The relation assumes Markovian exponential decay and separates energy relaxation from pure dephasing. It is not automatically valid for non-exponential or strongly correlated noise.

A phase-damping channel can be represented by Kraus operators such as E₀ = √((1+λ)/2) I and E₁ = √((1−λ)/2) Z, where λ is the remaining coherence factor, usually λ = e^(−t/T_φ).

3. Gate, measurement, and reset errors

Gate error

A gate error is the difference between the implemented quantum operation and the intended operation. It can arise from pulse-amplitude error, detuning, finite pulse duration, leakage outside the computational subspace, calibration drift, crosstalk, and control noise.

A practical model inserts a noise channel before or after an ideal gate:

This is an approximation. Real errors may depend on the gate, qubit state, neighbouring operations, time, and history. A two-qubit gate often has a substantially larger error than a single-qubit gate because it requires interaction between physical qubits.

The average gate fidelity compares an implemented channel with a target unitary. For a dimension-d system, one commonly used relation is

where F_e is entanglement fidelity under a specified convention. Reported gate error may mean 1 − F_avg, a randomized-benchmarking error per Clifford, or another metric. These quantities should not be treated as interchangeable without checking definitions.

Readout error

Readout error occurs when the classical result differs from the qubit state immediately before measurement. For one qubit, define assignment probabilities

The columns correspond to prepared states and the rows to reported results. Each column sums to one. If the true probability vector is p_true and the observed vector is p_obs, then

For independent qubits, a multi-qubit readout matrix may be approximated by a tensor product of single-qubit matrices. This assumption fails when measurement errors are correlated.

Readout mitigation estimates M from calibration experiments and attempts to invert or regularize the relation. Direct inversion can amplify statistical noise, especially when M is ill-conditioned. Mitigation does not recreate lost quantum information; it modifies estimates using a calibrated error model.

Reset errors

Reset is not always perfect. A qubit intended to be prepared in |0⟩ may retain excitation, and a reset operation may introduce additional dephasing or leakage. Reset errors matter in mid-circuit measurement, dynamic circuits, iterative algorithms, and error-correction routines.

4. From calibration data to model parameters

Calibration experiments estimate physical parameters. A relaxation experiment prepares |1⟩, waits for different durations, and measures the excited-state population. Fitting P₁(t) = A e^(−t/T₁) + B allows for preparation and readout offsets A and B. A simple idealized fit has A = 1 and B = 0.

A dephasing or Ramsey experiment typically applies a superposition, waits, and measures interference. The decay envelope estimates T₂^*, which includes both irreversible dephasing and slowly varying frequency fluctuations. An echo sequence can refocus some low-frequency noise and yield a different time, often denoted T₂^echo. Therefore T₂^* and T₂ should not be assumed identical.

Calibration data should be interpreted with uncertainty. If n independent shots produce k occurrences of an event, the estimated probability is hat p = k/n. Under a binomial model, its standard error is approximately

This estimate assumes independent identically distributed shots. Drift, temporal correlation, and changing calibration conditions violate that assumption. A fit with a small statistical error can still have substantial systematic error if the model is inappropriate.

Useful calibration questions include:

• What physical experiment produced the parameter?
• Over what time interval and temperature was it measured?
• Is the error stochastic, coherent, leakage-related, or correlated?
• Does the parameter apply to the exact gate and qubit pair being simulated?
• How recently was it calibrated?

5. Simulating noisy circuits

A circuit simulator may represent noise in several ways.

State-vector simulation is efficient for ideal pure states but cannot directly represent general mixed-state noise. It can approximate some channels by sampling random error trajectories.

Density-matrix simulation applies channels directly and represents mixed states exactly within the chosen model. For n qubits, the density matrix has 2ⁿ × 2ⁿ entries, so memory and computation grow rapidly.

Quantum trajectories sample individual stochastic outcomes using Kraus operators. Averaging many trajectories approximates density-matrix evolution and can be more efficient for some sparse processes, but introduces sampling uncertainty.

Pauli-noise simulation is often efficient because Pauli errors can be sampled classically. It is particularly useful for approximate studies, but coherent errors and non-Pauli effects may be misrepresented.

A practical noisy-simulation workflow is:

1. Define the ideal circuit and its expected observable.
2. Identify the hardware parameters relevant to each operation.
3. Select a model: for example, relaxation during gate durations, dephasing during idle periods, readout assignment error, and calibrated gate error.
4. State parameter conventions and assumptions.
5. Run enough shots or trajectories to separate physical effects from sampling fluctuations.
6. Compare ideal, noisy, and experimentally observed distributions.
7. Perform sensitivity analysis by varying uncertain parameters.
8. Record the simulator version, calibration snapshot, random seed, and noise assumptions.

Worked example: relaxation during a circuit

Suppose a qubit begins in |1⟩, has T₁ = 80 μs, and remains exposed to relaxation for t = 8 μs. The decay probability is

Thus the probability of still measuring the excited state, ignoring readout error and all other processes, is

The probability of relaxation is approximately 9.52%. If the measurement has a false-zero probability for |1⟩ of 3%, the observed probability of reporting 1 is approximately

assuming no other readout mechanism and independent relaxation and readout errors. The example illustrates why a measured error cannot automatically be attributed to the gate alone.

6. Topology, connectivity, and correlated effects

A device topology specifies which physical qubits can interact directly. If two logical qubits are not adjacent, a compiler may insert SWAP operations. A SWAP can require several two-qubit gates, increasing exposure to gate error, relaxation, and crosstalk. Consequently, two logically equivalent circuits can have different success probabilities after compilation.

Topology affects noise in at least four ways:

1. Routing overhead: additional gates and longer execution time.
2. Unequal calibration: different qubits and couplers have different error rates.
3. Spectator effects: an idle qubit may shift frequency or experience unwanted interactions when neighbouring qubits are driven.
4. Crosstalk: simultaneous operations can alter one another’s error rates.

Independent noise models assume that the total channel factors into local channels. This is computationally convenient but can fail for correlated dephasing, common control lines, measurement crosstalk, and two-qubit coherent errors. When correlation matters, a model may need joint Kraus operators, correlated Pauli terms, conditional error probabilities, or experimentally measured multi-qubit distributions.

A good compilation strategy therefore minimizes more than gate count. It may prefer qubits with longer coherence times, couplers with lower two-qubit error, shorter schedules, fewer idle intervals, and operations that avoid known crosstalk combinations.

7. Illustrative applied case: selecting a circuit layout

Background. A diploma-level laboratory team runs a three-qubit entangling circuit on a device whose connectivity is a line: qubits A–B–C. The intended circuit requires interactions between A and B and between A and C.

Problem. The A–C interaction is not direct. The compiler can either route through B using SWAP operations or remap logical qubits so that the most demanding interaction uses the better calibrated edge.

Analysis. Calibration shows that edge A–B has a two-qubit error of 1.2%, edge B–C has an error of 2.8%, and single-qubit errors are below 0.2%. The coherence times are also unequal: B has a shorter T₁ than A and C. A routing choice using B repeatedly may therefore incur both additional two-qubit error and extra idle-related relaxation.

Approach. The team compiles two layouts, simulates both using the same relaxation, dephasing, gate, and readout models, and runs each circuit with the same number of shots. It compares the target bit-string probability, not merely the total number of gates. It also repeats the experiment at two calibration times to detect drift.

Outcome. The layout that uses the lower-error edge for the dominant entangling operation produces the higher target probability, even though both layouts have similar nominal gate counts. The simulation predicts the ranking but underestimates the observed error because it omits correlated crosstalk during simultaneous pulses.

Learning points. Topology is part of the noise model. Calibration values must be associated with physical qubits and couplers. Agreement between simulation and experiment should be assessed at the level of distributions and uncertainties. A discrepancy is evidence for model limitations, not automatically evidence that the circuit is incorrectly programmed.

8. Limitations, governance, and best practices

Noise models are abstractions. A Markovian channel assumes that the future depends on the present state rather than the full history. Real devices can exhibit non-Markovian memory, drift, leakage, non-exponential decay, and temperature-dependent behaviour. A channel calibrated yesterday may not represent the device today.

Best practices include:

• Use the simplest model that answers the engineering question, then test whether added complexity changes the conclusion.
• Keep physical error types separate during diagnosis before combining them.
• Distinguish coherent errors, stochastic errors, leakage, and readout errors.
• Report parameter definitions, units, calibration time, confidence intervals, and shot counts.
• Validate simulators against analytically solvable circuits.
• Compare multiple noise models rather than presenting one prediction as exact.
• Use independent validation data when fitting or tuning a model.
• Avoid claiming that error mitigation produces fault-tolerant accuracy.

There are also governance issues. Calibration data may reveal device weaknesses and should be versioned and access-controlled where appropriate. Reproducible research requires recording hardware configuration, software versions, compiler settings, random seeds, and calibration snapshots. When noisy results inform scientific or industrial decisions, uncertainty and model risk must be communicated explicitly.

Contemporary work increasingly addresses leakage-aware simulation, pulse-level modelling, correlated noise, randomized compiling, dynamical decoupling, error mitigation, and early quantum error correction. These techniques reduce or characterize errors but do not eliminate the need to understand the underlying physical noise.

Chapter Summary

Open quantum systems interact with environments, so unitary state-vector evolution is insufficient for realistic computation. Density matrices represent mixed states, while completely positive trace-preserving channels describe physically valid state transformations. Kraus operators provide one practical representation of such channels.

Bit flips, phase flips, depolarization, amplitude damping, and dephasing represent different physical effects. Relaxation is characterized by T₁, while transverse coherence involves T₂ and, under suitable exponential assumptions, 1/T₂ = 1/(2T₁) + 1/T_φ. Gate errors, readout errors, reset errors, leakage, and crosstalk must be distinguished because they require different calibration and mitigation strategies.

Noisy-circuit simulation may use density matrices, trajectories, or stochastic Pauli approximations. Calibration data must be interpreted with attention to parameter definitions, uncertainty, drift, and correlations. Device topology influences routing overhead, exposure time, crosstalk, and the choice of physical qubits. Reliable professional practice combines analytical checks, calibrated models, experimental validation, and explicit reporting of assumptions.

Key Terms

Quantum channel: A physically valid map that transforms quantum states.

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

Density matrix: A positive, Hermitian, unit-trace operator describing pure or mixed states.

Complete positivity: The requirement that a map remains positive when extended to an entangled ancillary system.

Trace preservation: The condition that total probability remains one.

Bit flip: An error represented by the Pauli X operation.

Phase flip: An error represented by the Pauli Z operation.

Depolarizing noise: Noise that contracts the Bloch vector toward the maximally mixed state.

Amplitude damping: A channel modelling energy loss from an excited state.

Dephasing: Loss of phase coherence without necessarily changing populations.

T₁: Energy-relaxation time.

T₂: Transverse-coherence time.

Readout assignment error: Reporting a classical result different from the pre-measurement state.

Gate error: Deviation between an implemented and intended operation.

Crosstalk: Unwanted influence of one operation or qubit on another.

Leakage: Population leaving the computational subspace.

Quantum trajectory: A stochastic realization of channel evolution using Kraus outcomes.

Error mitigation: Methods that reduce bias in estimates without fully correcting the quantum state.

Calibration drift: Time-dependent change in device parameters.

Device topology: The connectivity pattern of physically interacting qubits.

Review Questions

Short Answer Questions

1. Why is a density matrix needed for general open-system evolution?
2. State the completeness condition for Kraus operators.
3. What physical process does amplitude damping model?
4. Distinguish T₁ from T₂.
5. Why can readout mitigation amplify statistical noise?

Long Answer Questions

1. Explain the operator-sum representation of a quantum channel and discuss complete positivity and trace preservation.
2. Compare bit-flip, phase-flip, depolarizing, amplitude-damping, and dephasing channels.
3. Describe how relaxation and dephasing calibration experiments estimate device parameters.
4. Compare density-matrix, quantum-trajectory, and Pauli-noise simulation methods.
5. Explain how device topology and crosstalk affect circuit compilation and observed accuracy.

Analytical / Application Questions

1. A qubit has T₁ = 100 μs and is exposed for 20 μs. Calculate its relaxation probability, assuming exponential decay.
2. A readout matrix and observed result frequencies are known. Explain how to estimate the true probability vector and why regularization may be required.
3. Design a noise-model validation experiment that separates gate error, relaxation, and readout error. State the controls, measurements, and assumptions.

Multiple Choice Questions with Answers

Question 1

Which condition must Kraus operators satisfy for a trace-preserving channel?

A. ∑k EkEk† = 0

B. ∑k Ek†Ek = I

C. ∑k Ek = I

D. Ek†Ek = ρ for every k

Correct Answer: B. ∑k Ek†Ek = I

Question 2

What does amplitude damping primarily model?

A. Energy loss from an excited state

B. A deterministic phase rotation only

C. Classical sampling noise only

D. A change in qubit connectivity

Correct Answer: A. Energy loss from an excited state

Question 3

For an initially excited qubit with T1 = 80 μs and exposure time 8 μs, what is the approximate relaxation probability?

A. 0.0100

B. 0.0488

C. 0.0952

D. 0.8000

Correct Answer: C. 0.0952

Question 4

Under the exponential-decay assumptions in the chapter, which relation connects T2, T1, and pure-dephasing time Tφ?

A. T2 = T1 + Tφ

B. 1/T2 = 1/(2T1) + 1/Tφ

C. 1/T2 = 2T1 + Tφ

D. T2 = 2T1Tφ

Correct Answer: B. 1/T2 = 1/(2T1) + 1/Tφ

Question 5

What is a principal effect of a phase flip?

A. It changes |1⟩ into |0⟩ with certainty

B. It changes relative phase while preserving computational-basis populations

C. It always removes the qubit from the computational subspace

D. It increases T1

Correct Answer: B. It changes relative phase while preserving computational-basis populations

Question 6

In a readout model p_obs = M p_true, what does M represent?

A. A unitary gate applied before measurement

B. A matrix of conditional assignment probabilities

C. The density matrix of the environment

D. The number of circuit shots

Correct Answer: B. A matrix of conditional assignment probabilities

Question 7

Why can routing reduce the reliability of a quantum circuit?

A. Routing removes all idle time

B. Routing may add SWAP gates and increase exposure to errors

C. Routing converts all gates into measurements

D. Routing guarantees perfect connectivity

Correct Answer: B. Routing may add SWAP gates and increase exposure to errors

Question 8

Which simulator representation directly supports general mixed-state channel evolution?

A. A density matrix

B. A classical truth table only

C. A single ideal state vector without sampling

D. A circuit diagram without parameters

Correct Answer: A. A density matrix

Question 9

What is crosstalk?

A. The desired evolution of an isolated qubit

B. Unwanted influence between simultaneous operations or neighbouring qubits

C. A method for measuring T1 only

D. A requirement that all channels be unital

Correct Answer: B. Unwanted influence between simultaneous operations or neighbouring qubits

Question 10

Why should reported gate-error metrics be interpreted carefully?

A. All gate-error definitions are mathematically identical

B. They may refer to different metrics such as average infidelity or benchmarking error

C. Gate errors cannot depend on calibration

D. Gate errors are always caused by readout

Correct Answer: B. They may refer to different metrics such as average infidelity or benchmarking error