8. Classical Simulation of Quantum Systems and Circuits

Introduction and Context

Quantum computation is physically realized through quantum devices, but much of its theory, algorithm design, verification, and software development depends on classical simulation. A simulator represents quantum states, operations, measurements, noise, or physical dynamics using conventional computational resources. Simulation therefore serves two complementary purposes: it is a scientific instrument for studying quantum phenomena and an engineering instrument for developing programs before access to quantum hardware.

The central difficulty is dimensional growth. An n-qubit pure state is represented by a vector of 2^n complex amplitudes. A general mixed state requires a 2^n × 2^n density matrix, containing 4^n complex entries before exploiting structure. This exponential scaling does not imply that every quantum computation is equally difficult to simulate. Product states, stabilizer states, low-entanglement states, sparse Hamiltonians, shallow circuits, and structured noise can often be simulated efficiently. The practical question is consequently not simply whether a simulator is “classical” or “quantum,” but which representation preserves the relevant structure at acceptable cost and error.

This chapter develops a comparative framework for state-vector, density-matrix, stabilizer, tensor-network, path-integral, Monte Carlo, and distributed high-performance-computing (HPC) and GPU methods. It emphasizes representation, computational complexity, approximation, numerical stability, validation, and the relationship between algorithmic structure and hardware resources.

Learning Objectives

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

1. Explain why the classical simulation of quantum systems is computationally challenging.

2. Select an appropriate simulation representation for a circuit, noise model, or physical Hamiltonian.

3. Compare state-vector, density-matrix, stabilizer, tensor-network, path-integral, and Monte Carlo methods.

4. Analyze memory, time, communication, entanglement, variance, and numerical-stability constraints.

5. Design validation procedures for simulator outputs.

6. Explain how distributed-memory systems, GPUs, and hybrid workflows extend simulation capability.

7. Evaluate professional, scientific, and governance risks associated with simulation claims.

The Representation Determines the Simulation Problem

A simulator begins by choosing what object will be stored and evolved. For a pure state,

|ψ⟩ = Σx∈{0,1}ⁿ αx|x⟩,

where the amplitudes satisfy Σx|αx|² = 1. A quantum gate U transforms the state as |ψ′⟩ = U|ψ⟩. In a computational-basis state-vector simulator, the principal data structure is an array of 2^n complex numbers.

A noisy or statistically uncertain state is represented by a density operator ρ. It evolves under a unitary operation according to ρ′ = UρU†, or under a quantum channel according to ρ′ = Σk EkρEk†. The density matrix captures coherence, classical uncertainty, and mixedness, but its quadratic increase relative to the state-vector dimension makes direct simulation substantially more expensive.

Other methods avoid storing all amplitudes. Stabilizer simulation stores a compact description of states generated by Clifford operations, while tensor networks factor a high-dimensional object into smaller tensors connected according to a graph. Path-integral methods sum contributions from computational histories, and Monte Carlo methods estimate observables through samples rather than explicitly representing the complete state. Each method exchanges generality for structure, memory, variance, or approximation.

Three distinctions are particularly important. First, exact representation differs from approximate representation. Second, simulating a circuit differs from estimating a physical observable. Third, the number of qubits alone is an inadequate predictor of difficulty. Entanglement, circuit depth, interaction graph, noise rank, sign cancellations, and the desired output all influence cost.

State-Vector Simulation: The General-Purpose Baseline

State-vector simulation is the most direct method for pure-state circuits. A one-qubit gate acting on qubit q updates pairs of amplitudes whose binary indices differ only at bit q. A two-qubit gate updates groups of four amplitudes. Applying a gate therefore takes time proportional to the number of amplitudes, generally O(2^n) for each nontrivial operation, while memory is O(2^n).

The method is attractive because it is conceptually transparent and supports arbitrary unitary gates, measurements, and many circuit transformations. It is often the reference implementation against which specialized simulators are tested. Measurement can be performed by computing probabilities, sampling an outcome, and renormalizing the post-measurement state. Repeated shots may be generated efficiently once the state or relevant marginal distributions are available.

The cost is dominated by memory bandwidth and capacity rather than arithmetic alone. With double-precision complex amplitudes, each amplitude requires 16 bytes. A state with n qubits therefore requires approximately 16·2^n bytes before workspace and communication overhead. At larger sizes, the state is partitioned among nodes, and gates whose target bits are distributed across nodes require data exchange.

Numerical issues arise even in apparently simple simulations. Repeated operations can accumulate floating-point error, causing norm drift or small violations of Hermiticity after transformations. Renormalization may control norm error but can conceal a faulty kernel. A robust implementation records norm, probability conservation, expectation values, and, where possible, reversibility tests. State-vector simulation is exact only relative to floating-point arithmetic and the specified model; it does not reproduce hardware imperfections unless those are explicitly modeled.

Density-Matrix and Quantum-Trajectory Methods

Density-matrix simulation is the natural general method for open quantum systems. Direct propagation stores 4^n complex matrix elements and commonly requires O(4^n) memory and O(4^n) work per simple channel application, with larger costs for general operations. It supports decoherence, correlated noise, thermal states, leakage models, and arbitrary completely positive trace-preserving maps.

For a Markovian master equation,

dρ/dt = −i[H,ρ] + Σk (LkρLk† − 1/2{Lk†Lk,ρ}),

one may integrate the equation using matrix-exponential, Runge–Kutta, Krylov, or specialized sparse methods. The appropriate integrator depends on stiffness, time dependence, sparsity, and the desired error tolerance. Positivity is a central concern: a numerical approximation can preserve trace while producing a matrix with small negative eigenvalues. Such artifacts must be distinguished from physically meaningful populations.

Quantum trajectories provide a compromise. Instead of evolving ρ directly, one evolves many stochastic pure-state trajectories under an effective non-Hermitian Hamiltonian and quantum jumps. Observable estimates are averages over trajectories. The method can reduce memory from density-matrix scale to state-vector scale, but introduces statistical error. Its cost is governed by the number of trajectories required to achieve a desired confidence interval, particularly when rare jumps or weak signals are involved.

Thus, density matrices offer deterministic ensemble evolution at severe dimensional cost, whereas trajectories trade deterministic representation for sampling variance. Neither dominates universally. A low-noise system with many observables may favor trajectories; a small system requiring precise correlated expectations may favor direct density-matrix evolution.

Stabilizer Simulation and the Boundary of Efficient Classical Computation

The stabilizer formalism represents a restricted but important family of quantum states. Clifford gates—such as Pauli operations, controlled-NOT, and Hadamard—map Pauli operators to Pauli operators under conjugation. Gottesman–Knill-type methods exploit this closure to simulate circuits using polynomial classical resources in the number of qubits and gates.

A stabilizer tableau records generators of the stabilizer group and updates them through algebraic row operations. Measurement of Pauli observables can also be handled efficiently. This makes stabilizer simulation valuable for quantum error-correction circuits, syndrome extraction, Clifford benchmarking, and large circuits containing predominantly Clifford operations.

The restriction is fundamental. A non-Clifford gate, such as a generic rotation or T gate, generally takes a stabilizer state outside the efficiently represented class. Several extensions address this boundary. Stabilizer-rank methods express a non-stabilizer state as a linear combination of stabilizer states; the number of terms controls computational cost. Quasiprobability decompositions similarly represent a target operation as a signed combination of efficiently simulable operations, but sampling variance can grow rapidly.

Stabilizer methods illustrate a broader principle: computational tractability often follows from algebraic structure, not from a small physical system. They can simulate circuits with thousands or more qubits when the circuit remains within the tractable family, yet become inefficient when non-Clifford content and entanglement increase.

Tensor Networks and Entanglement-Aware Simulation

A tensor network represents a quantum state or circuit as tensors connected by indices. A circuit can be viewed as a network whose contraction produces an amplitude, expectation value, or probability. Matrix product states (MPS) are particularly effective for one-dimensional systems with limited entanglement. The bond dimension χ controls the amount of correlation retained, with storage commonly scaling as O(nχ²) for fixed local dimension.

The von Neumann entropy across a bipartition provides a conceptual guide. If entanglement entropy remains modest, a relatively small bond dimension can approximate the state. If entropy grows substantially, χ must increase, and the simulation may approach exponential cost. Two-dimensional geometries, long-range interactions, random deep circuits, and volume-law entanglement are therefore challenging.

Tensor-network simulation involves two different optimization problems: choosing a network representation and choosing a contraction order. Contraction ordering can change cost by many orders of magnitude. Practical systems use graph heuristics, slicing, truncation, symmetry, and parallel contraction. Truncating singular values after a decomposition reduces bond dimension but introduces approximation error. The discarded weight provides a local diagnostic, not a complete guarantee of global observable accuracy.

Tensor networks can simulate full states, selected amplitudes, or expectation values. They are particularly useful for variational algorithms, low-dimensional many-body physics, time evolution, and circuits with favorable geometry. Their failure is informative: rapid bond-dimension growth indicates that the chosen representation no longer captures the computation economically.

Path Integrals, Feynman Sums, and Monte Carlo Estimation

A circuit amplitude can be written as a sum over computational paths. Each path contributes a product of local transition amplitudes, and interference arises because contributions may have different phases. Path-integral simulation can exploit shallow depth, limited light cones, or a small number of measured outputs without storing the complete state.

The difficulty is cancellation. A large number of paths may contribute terms whose sum is small, requiring high precision or sophisticated contraction and sampling strategies. In classical statistical mechanics, Monte Carlo methods often use nonnegative probability weights. Quantum amplitudes and imaginary-time formulations may instead produce negative or complex weights. The resulting sign or phase problem can make variance grow exponentially and is a major barrier to generic classical simulation.

Monte Carlo methods estimate an observable O through samples x₁,…,xN. The standard error is approximately σO/√N when samples are sufficiently independent, although autocorrelation reduces the effective sample size. Confidence intervals must account for variance, burn-in, thermalization, and correlation. Importance sampling can focus effort on influential configurations, while Markov-chain methods generate samples from distributions that are difficult to sample directly.

For real-time quantum dynamics, stochastic approaches may involve complex weights or stochastic unravelings. For imaginary-time ground-state calculations, projector Monte Carlo and path-integral methods may be effective when the sign problem is absent or controlled. A reported Monte Carlo estimate without an uncertainty analysis is incomplete: the estimate and its error bar are inseparable scientific outputs.

Choosing Among Simulation Methods

Method selection should begin with the computational question rather than the simulator brand. The following decision logic is useful:

• Use a state vector when the system is pure, the qubit count is moderate, and arbitrary gates or complete output distributions are required.
• Use a density matrix when general noise, mixed states, or open-system dynamics are central and the system is small enough for quadratic state dimension.
• Use stabilizer methods when Clifford structure dominates, especially in error-correction or syndrome circuits.
• Use tensor networks when geometry and entanglement are favorable, and approximation through controlled truncation is acceptable.
• Use path-sum or amplitude methods when only selected outputs, shallow circuits, or restricted light cones are required.
• Use Monte Carlo or trajectory methods when observables can be estimated from samples and variance is manageable.
• Use distributed HPC and GPUs when the selected representation is suitable but exceeds the memory or throughput of a single workstation.

A useful evaluation matrix includes: representation size, gate or timestep cost, communication volume, approximation error, sampling variance, entanglement growth, numerical stability, and validation options. The method should be selected against the observable and accuracy requirement. Simulating an entire state is unnecessary if the research question concerns one local expectation value; conversely, estimating one observable is insufficient when a complete output distribution is needed.

Distributed HPC and GPU Architectures

Large simulations commonly distribute amplitudes or tensors across nodes. A one-dimensional partition assigns contiguous blocks of a state vector to processes. Gates acting on local index bits require no communication, while gates acting on distributed bits require exchanges, often implemented through collective communication or pairwise remapping. Communication latency and network bandwidth can dominate arithmetic at scale.

Two-dimensional or hierarchical decompositions can improve communication patterns, particularly for tensor contractions or density matrices. Checkpointing is important because large simulations may run for hours or days; however, checkpoint files can themselves be enormous. Reproducibility requires recording partitioning, precision, random seeds, software versions, gate ordering, and numerical tolerances.

GPUs are effective because gate application and tensor contraction expose substantial parallelism. Their performance depends on coalesced memory access, kernel fusion, occupancy, and host-device transfer avoidance. A simulation may be computationally intensive but still perform poorly if each gate launches a small kernel or requires repeated synchronization. Mixed precision can increase throughput, but it should be adopted only after measuring its effect on norms, expectation values, and long-time stability.

Hybrid CPU–GPU and multi-GPU systems introduce additional concerns: load imbalance, topology-aware communication, asynchronous execution, and fault recovery. Hardware acceleration does not change asymptotic complexity. It changes the practical boundary by increasing available memory bandwidth and parallel throughput.

Validation, Verification, and Numerical Trustworthiness

A simulator must be validated at several levels. Verification asks whether the implementation correctly solves the specified mathematical model. Validation asks whether the model and approximation represent the physical or computational target adequately.

Basic verification tests include:

1. Norm preservation for unitary pure-state evolution.

2. Trace preservation and Hermiticity for density-matrix evolution.

3. Positivity checks for density operators and channels.

4. Agreement with analytically solvable one- and two-qubit circuits.

5. Equivalence under circuit identities, such as applying a gate and its inverse.

6. Cross-validation against an independent simulator or representation.

7. Convergence studies over timestep, bond dimension, sample count, and precision.

Randomized testing is useful, but random tests should not replace carefully designed edge cases. For stochastic simulation, repeated independent runs should produce confidence intervals whose empirical coverage is plausible. For tensor networks, increasing bond dimension should lead to convergence of the reported observable; a decreasing truncation threshold should produce a consistent trend. For distributed programs, results should be compared across process counts because race conditions and reduction-order effects may otherwise remain hidden.

Floating-point arithmetic is non-associative. Consequently, parallel reductions can produce slightly different answers even when the algorithm is correct. The relevant question is whether discrepancies are bounded by a stated tolerance and whether they affect scientific conclusions. Reproducibility should therefore be defined quantitatively, not as an unrealistic demand for bitwise identity in every execution.

Illustrative Case Study: Selecting a Simulator for a Noisy Surface-Code Experiment

Background

Consider an illustrative research group studying a small surface-code memory. The circuit contains repeated stabilizer-measurement rounds, mostly Clifford gates, preparation and measurement operations, and a configurable model of single-qubit and two-qubit noise. The group wants logical-error rates across many physical-error probabilities and also wishes to inspect selected correlated observables.

Situation and Problem

A full density-matrix simulation is immediately impractical because the state dimension grows quadratically relative to a state vector. A generic state-vector simulator can represent the ideal circuit but would require repeating the computation for many noise realizations or explicitly modeling channels. The researchers must choose a method that supports large numbers of qubits, repeated rounds, and statistical estimation.

Analysis and Strategy

The circuit is predominantly Clifford. The group therefore begins with a stabilizer simulator for the ideal and Pauli-noise portions. Pauli errors can be sampled as classical fault configurations, allowing many trials to be run efficiently. To study a small non-Clifford calibration operation introduced in one variant, the researchers use a stabilizer-rank or direct state-vector simulation on a reduced subsystem and compare the result with the main stabilizer workflow.

They define the logical-error event before running experiments, fix random seeds for development tests, and use independent seeds for production trials. For each physical-error rate, they report the number of trials, estimated logical-error probability, and confidence interval. Small-distance instances are cross-checked with density-matrix simulation, while circuit identities and noiseless stabilizer invariants are tested automatically.

Outcome

The stabilizer-based workflow permits broad parameter sweeps that would be infeasible with density matrices. The reduced direct simulations quantify the effect of the non-Clifford component, while cross-validation identifies the range in which the simplified Pauli-noise model remains adequate. The researchers do not claim that the simulation predicts all hardware behavior; they explicitly identify leakage, non-Pauli correlations, calibration drift, and measurement imperfections as unmodeled factors.

Learning Points

The case demonstrates that method selection should follow circuit structure and research objectives. It also shows why a fast simulator can still produce weak science if the noise model, uncertainty analysis, and validation plan are poorly specified. Hybrid simulation is often more defensible than forcing one universal representation onto every component.

Challenges, Risks, and Common Mistakes

A frequent mistake is to compare simulators solely by maximum qubit count. A stabilizer simulator may handle many qubits but not a highly non-Clifford circuit; a tensor-network simulator may handle a large one-dimensional system but fail after entanglement growth; a Monte Carlo method may store little state but require an impractical number of samples.

Another mistake is to confuse floating-point precision with model accuracy. Double precision cannot correct an incomplete Hamiltonian or an incorrect noise channel. Likewise, a small truncation error on each tensor does not automatically imply a small error in a long-time observable.

Common technical risks include memory exhaustion, communication bottlenecks, hidden synchronization, unstable differential-equation integration, biased Monte Carlo sampling, inadequate burn-in, insufficient trajectory counts, and accidental reuse of random streams. In density-matrix work, trace preservation alone is not enough; positivity and complete positivity of the implemented channel must also be considered.

Researchers should avoid reporting only favorable parameter regimes. A credible study describes failure boundaries, sensitivity to precision and discretization, and the conditions under which conclusions change. Benchmarking should include time-to-solution, peak memory, communication cost, energy or resource use where relevant, and accuracy—not merely raw throughput.

Ethical, Governance, and Professional Considerations

Simulation results can influence claims about quantum advantage, cryptographic capability, error-correction thresholds, and investment decisions. Such claims require clear separation between exact results, numerical approximations, and extrapolations beyond the validated regime. Reproducible artifacts, documented models, and disclosure of unavailable or proprietary components strengthen scientific accountability.

There is also a governance issue in benchmarking. Selecting a favorable simulator, precision, subset of outputs, or stopping criterion can create misleading comparisons. Performance reports should disclose hardware, software versions, compiler settings, circuit families, accuracy criteria, and preprocessing. In security-related applications, simulation should not be treated as evidence that a protocol is secure under hardware behavior not represented in the model.

Current Developments and Emerging Directions

Current research increasingly combines representations rather than treating them as mutually exclusive. Stabilizer decompositions, tensor-network contractions, trajectory sampling, and GPU kernels may be combined within one workflow. Automatic circuit slicing and contraction planning use graph optimization to reduce memory and communication. Symmetry-aware tensor networks and compressed representations exploit conserved quantities that would otherwise be discarded.

Differentiable simulators are being developed for variational algorithms and control, enabling gradients through state-vector, tensor-network, or stochastic computations. This creates new numerical concerns because gradient variance and truncation errors can distort optimization. Surrogate models and learned contraction strategies may improve performance, but they require independent validation and should not obscure the underlying approximation.

Quantum-classical co-simulation is also becoming important. Classical simulation can model control electronics, calibration routines, and noise, while quantum processors execute selected subcircuits. The resulting hybrid experiments require careful interface definitions, synchronization, and uncertainty propagation.

Best-Practice Workflow

A defensible simulation project generally proceeds as follows:

1. Define the observable, output distribution, accuracy target, and physical model.

2. Estimate representation size and identify exploitable structure.

3. Select a baseline simulator and an independent validation method.

4. Implement small analytically tractable tests before large experiments.

5. Measure memory, communication, runtime, and numerical diagnostics.

6. Perform convergence studies over relevant approximation parameters.

7. Quantify sampling uncertainty and distinguish it from systematic error.

8. Record software, hardware, precision, seeds, and model assumptions.

9. Report both successful regimes and failure boundaries.

10. Preserve code, input circuits, and metadata sufficient for reproduction.

Chapter Summary

Classical simulation of quantum systems is governed by the representation chosen for the state, dynamics, and observable. State-vector methods are general for pure circuits but scale exponentially in memory. Density matrices support mixed states and open-system channels at a still more severe cost, while quantum trajectories exchange memory demands for sampling variance. Stabilizer methods exploit Clifford structure, tensor networks exploit favorable entanglement and geometry, and path-integral or Monte Carlo methods estimate selected quantities through histories or samples. Distributed HPC and GPUs extend practical scale but do not remove asymptotic complexity.

Reliable simulation requires more than fast execution. It requires analysis of approximation, entanglement, communication, variance, numerical stability, and model adequacy. Verification through invariants and independent implementations must be complemented by convergence studies and uncertainty quantification. The most effective professional practice is often hybrid: select a representation for each structured part of the problem, validate it against a simpler or independent method, and state clearly where the conclusions cease to be trustworthy.

Key Terms

State vector: A complex-amplitude vector representing a pure quantum state.

Density matrix: An operator representing mixed states, statistical uncertainty, and open-system evolution.

Quantum channel: A completely positive trace-preserving map describing general quantum evolution.

Stabilizer state: A state characterized as the simultaneous eigenstate of a commuting Pauli group.

Clifford gate: A gate that maps Pauli operators to Pauli operators under conjugation.

Tensor network: A factorized representation of a high-dimensional tensor using connected lower-dimensional tensors.

Bond dimension: The dimension of an internal tensor-network index controlling retained correlations.

Entanglement entropy: A measure of quantum correlation across a bipartition.

Path integral: A formulation expressing an amplitude as a sum over possible computational or physical histories.

Monte Carlo method: A numerical method that estimates quantities using random or pseudorandom samples.

Quantum trajectory: A stochastic pure-state realization whose ensemble average reproduces open-system evolution.

Sign problem: The difficulty caused by negative or complex sampling weights and severe cancellation.

Truncation error: Error introduced by discarding small components, such as singular values, in an approximation.

Sampling variance: Statistical variation in an estimator caused by finite random samples.

Distributed simulation: Simulation whose data and computation are partitioned across multiple processors or nodes.

GPU acceleration: Use of graphics-processing units to execute massively parallel simulation kernels.

Numerical stability: The extent to which computed results remain controlled under rounding and perturbation.

Verification: Testing whether an implementation correctly solves its specified mathematical problem.

Validation: Assessing whether the mathematical model and approximation represent the intended physical or computational target.

Review Questions

Short Answer Questions

1. Why does a general n-qubit density matrix require substantially more storage than a state vector?

2. What structural property makes stabilizer simulation efficient for Clifford circuits?

3. How does bond dimension affect tensor-network simulation?

4. Why must Monte Carlo results include uncertainty estimates?

5. What is the difference between verification and validation?

Long Answer Questions

1. Compare state-vector and density-matrix simulation in terms of representation, complexity, applications, and numerical risks.

2. Explain how entanglement and circuit geometry determine the success or failure of tensor-network methods.

3. Discuss the advantages and limitations of stabilizer, stabilizer-rank, and quasiprobability approaches.

4. Analyze the role of distributed memory, communication, and GPUs in large-scale quantum simulation.

5. Develop a validation framework for a noisy quantum-circuit simulator.

Analytical / Application Questions

1. A circuit contains thousands of Clifford gates, a small number of T gates, and Pauli noise. Which simulation strategy would you select, and how would you validate it?

2. An MPS simulation shows rapid growth in bond dimension during time evolution. What does this indicate, and what alternatives or approximation controls should be considered?

3. A Monte Carlo estimate changes little when the sample count doubles, but its confidence interval remains large. What might explain this result, and what further diagnostics are appropriate?

22. Multiple Choice Questions with Answers

Question 1

For an n-qubit pure-state simulation using a conventional state vector, which resource scales exponentially with n?

A. The number of amplitudes, which is 2^n

B. The number of amplitudes, which is n²

C. The number of gates, which is always 2^n

D. The number of measurement outcomes, which is always n

Correct Answer: A. The number of amplitudes, which is 2^n

Question 2

Which simulation method is most naturally suited to a circuit composed almost entirely of Clifford gates?

A. Direct density-matrix propagation

B. Stabilizer simulation

C. Generic path-integral summation

D. Full tensor contraction without structure

Correct Answer: B. Stabilizer simulation

Question 3

What is the principal trade-off in quantum-trajectory simulation compared with direct density-matrix evolution?

A. It removes all numerical error

B. It uses state-vector-scale memory but introduces sampling variance

C. It requires no noise model

D. It is limited exclusively to Clifford gates

Correct Answer: B. It uses state-vector-scale memory but introduces sampling variance

Question 4

In an MPS simulation, rapid growth of bond dimension usually indicates that:

A. The state is becoming easier to represent exactly

B. Entanglement is increasing and the representation is becoming more costly

C. The circuit has become purely classical

D. Floating-point precision is automatically improving

Correct Answer: B. Entanglement is increasing and the representation is becoming more costly

Question 5

Why can Monte Carlo simulation of a quantum path integral become difficult even when individual paths are easy to generate?

A. All paths necessarily have identical positive weights

B. Destructive interference can cause sign or phase cancellations and large variance

C. Monte Carlo cannot estimate observables

D. Quantum systems have no measurable probabilities

Correct Answer: B. Destructive interference can cause sign or phase cancellations and large variance

Question 6

Which diagnostic is especially appropriate for direct density-matrix evolution?

A. Checking only the number of CPU cores

B. Checking trace preservation, Hermiticity, and positivity

C. Checking only tensor bond dimension

D. Checking whether every gate is Clifford

Correct Answer: B. Checking trace preservation, Hermiticity, and positivity

Question 7

In distributed state-vector simulation, why can a gate acting on a distributed qubit be expensive?

A. It may require communication or data exchange between nodes

B. It changes the number of qubits

C. It eliminates all parallelism

D. It always requires conversion to a density matrix

Correct Answer: A. It may require communication or data exchange between nodes

Question 8

A researcher needs only one local expectation value from a low-entanglement one-dimensional system. Which approach may avoid unnecessary computational cost?

A. A tensor-network method targeting the expectation value

B. A full density matrix regardless of system size

C. Enumerating every possible hardware fault

D. Storing every computational-basis history explicitly

Correct Answer: A. A tensor-network method targeting the expectation value

Question 9

What does a convergence study in a tensor-network simulation typically vary?

A. Bond dimension or truncation threshold

B. The meaning of the observable

C. The definition of a qubit

D. The number of Pauli operators in the theory

Correct Answer: A. Bond dimension or truncation threshold

Question 10

Which statement best distinguishes verification from validation?

A. Verification tests implementation correctness; validation tests adequacy for the intended target

B. Verification concerns hardware cost; validation concerns compiler syntax only

C. Verification is statistical sampling; validation is memory allocation

D. They are identical terms with no practical distinction

Correct Answer: A. Verification tests implementation correctness; validation tests adequacy for the intended target