Source-linked AI summary
Myths around quantum computation before full fault tolerance: What no-go theorems rule out and what they don't
Zoltán Zimborás, Bálint Koczor, Zoë Holmes, Elsi-Mari Borrelli, András Gilyén, Hsin-Yuan Huang, Zhenyu Cai, Antonio Acín, Leandro Aolita, Leonardo Banchi, Fernando G. S. L. Brandão, Daniel Cavalcanti, Toby Cubitt, Sergey N. Filippov, Guillermo García-Pérez, John Goold, Orsolya Kálmán, Elica Kyoseva, Matteo A. C. Rossi, Boris Sokolov, Ivano Tavernelli, Sabrina Maniscalco
TL;DR
The paper addresses how theoretical no-go results should inform, rather than foreclose, expectations for quantum computing before full fault tolerance. It uses a perspective-based synthesis of error mitigation, variational algorithms, and hardware evolution, concluding that finite practical regimes and near-term opportunities remain while important training and scalability limitations persist.
Problem
The paper examines misconceptions about the practical implications of no-go results, error mitigation, and variational quantum algorithms during the transition toward fault-tolerant quantum computing.
Method
The authors reevaluate theoretical results and prevailing beliefs while connecting error mitigation, variational methods, and the progression from NISQ to fault-tolerant architectures.
Results
The perspective concludes that exponential error-mitigation costs can still leave finite feasible regimes, while variational training faces severe barren-plateau and copy-complexity obstacles.
Takeaways & Limitations
Near-term quantum applications remain possible through hardware and circuit-design improvements, with error mitigation potentially extending into early error-corrected machines.
Takeaways & Limitations
Training unstructured quantum circuits at scale is constrained by barren plateaus, degraded precision scaling, copy complexity, and difficulty fitting many parameters into NISQ circuits.
Abstract
from arXiv · showhide
In this perspective article, we revisit and critically evaluate prevailing viewpoints on the capabilities and limitations of near-term quantum computing and its potential transition toward fully fault-tolerant quantum computing. We examine theoretical no-go results and their implications, addressing misconceptions about the practicality of quantum error mitigation techniques and variational quantum algorithms. By emphasizing the nuances of error scaling, circuit depth, and algorithmic feasibility, we highlight viable near-term applications and synergies between error mitigation and early fault-tolerant architectures. Our discussion explores strategies for addressing current challenges, such as barren plateaus in variational circuits and the integration of quantum error mitigation and quantum error correction techniques. We aim to underscore the importance of continued innovation in hardware and algorithmic design to bridge the gap between theoretical potential and practical utility, paving the way for meaningful quantum advantage in the era of late noisy intermediate scale and early fault-tolerant quantum devices.
I. INTRODUCTION
The article argues that theoretical no-go results should not be treated as definitive statements about practical near-term quantum computing. It presents quantum hardware as evolving continuously from early NISQ through late NISQ and fault-tolerant stages, while examining myths about error mitigation and variational algorithms.
- Theoretical limitations on NISQ-friendly methods have tempered earlier unrealistic expectations about near-term quantum computation.
- No-go results often concern asymptotic regimes and can be misinterpreted as definitive claims about practical performance.
- Quantum hardware is expected to transition through early NISQ, late NISQ, early fault tolerance, and full fault tolerance.
- The perspective reevaluates myths about error mitigation and variational algorithms while identifying potential near-term applications and synergies with early error-corrected machines.
A. Utility of Quantum Error Mitigation
Quantum error mitigation operates directly on physical qubits rather than using encoded logical qubits and correction protocols. It reduces noise-induced bias in expectation-value estimates, but increases estimator variance and therefore sampling requirements.
- Quantum error mitigation works directly on physical qubits without encoding information into logical qubits.
- QEM targets bias introduced by noise in expectation-value estimates obtained from repeated circuit measurements.
- Reducing bias through mitigation trades it for increased estimator variance and more samples for accurate results.
1. Exponential scaling of error mitigation methods
Error mitigation has exponential sampling costs as circuit size grows, but the gate-error prefactor in the exponent can leave a finite feasible regime. Theoretical lower bounds align with the exponential overheads of leading mitigation methods.
- Under suitable noise models, error mitigation requires exponentially many samples because noisy outputs become difficult to distinguish from maximally mixed states.
- For brickwork circuits with N qubits and depth L, the required sample overhead has an Ω(exp(εNL)) lower bound under layer-wise local depolarizing noise.
- State-of-the-art mitigation methods have sampling costs of p(εNL) exp(εBNL), matching the exponential character of the lower bound.
- A small gate-error rate ε can permit circuits with NL proportional to ε^-1 to remain feasible despite exponential scaling.
- The same mitigation framework may remain relevant on early error-corrected machines, where ε becomes a logical error rate reduced by increasing code distance.
2. Feasible circuit sizes in near-term quantum computing
Pre-fault-tolerant devices can support useful circuit sizes when sampling budgets, gate-error rates, and circuit complexity are jointly constrained. The article argues that practical advantage remains possible for selected applications, but has not yet been demonstrated.
- Current and near-future two-qubit gate errors motivate estimating useful circuit sizes from feasible sampling overheads.
- For superconducting devices, 1000–5000 circuit layer operations per second make 10^6–10^7 samples feasible for depths of 100–1000.
- A practical sampling budget of approximately e^10 corresponds to an average of about 10 circuit errors and NL proportional to ε^-1.
- A 100-qubit circuit with depth beyond 10^5 would require error rates below 10^-6, beyond current hardware-provider roadmaps.
- Quantum dynamics and many-body experiments may achieve near-term practical advantage if circuit sizes are minimized and error mitigation is optimized.
- Pre-fault-tolerant circuit sizes may enable useful applications, but practical quantum advantage remains to be demonstrated.
3. Path towards fault tolerance
Early fault-tolerant devices will retain non-negligible logical errors, so quantum error mitigation can complement error correction throughout the transition toward full fault tolerance.
- Integrating QEM and QEC: Existing quantum error mitigation methods can be extended from physical to logical errors by replacing physical operations and errors with logical counterparts.Logical PEC faces harder noise characterization but can benefit from known logical error models and virtual Pauli-frame updates.
- Integrating QEM and QEC: Purification methods avoid noise characterization, while error-corrected zero-noise extrapolation can extrapolate toward infinite code distances.The paper identifies these as mitigation methods with additional benefits in logical contexts.
- Beyond device noise: Quantum error mitigation can address compilation and algorithmic errors that quantum error correction cannot remove, including rotation-synthesis and finite-Trotter-step errors.PEC can synthesize continuous rotations exactly on average, while extrapolation can suppress or eliminate finite-step effects.
- Integrating QEM and QEC: QEM and QEC can also be integrated through noise-entanglement circuits that use encoded registers and parity checks to reveal errors in unencoded registers.The proposed construction can combine bit-flip and phase-flip codes to obtain the error-correction power of a surface code.
- Overall verdict: Quantum error mitigation is expected to remain a major enabler across successive hardware generations until full fault tolerance is reached.The paper presents hardware evolution as continuous rather than as a sharp replacement of QEM by QEC.
B. Variational quantum algorithms
Variational quantum computing combines flexible, hardware-adaptable optimization with heuristic behavior that makes its resource requirements and scalability difficult to predict.
- Strengths: Variational quantum computing is flexible and adaptable to near-term hardware, making it applicable to a broad range of problems.These strengths are inherited from classical optimization-based algorithms.
- Limitations: Variational quantum algorithms generally lack guarantees on resource requirements, unlike more conventional quantum algorithms.The absence of immediately available hardware for testing further complicates predictions about scaling to interesting problem sizes.
1. Fundamental limitations of variational quantum algorithms
Variational quantum algorithms face barriers from difficult loss landscapes, barren plateaus, and training overheads, but problem-inspired models and non-variational components may preserve viable regimes.
- Barren plateaus: Barren plateaus make gradients vanish for most randomly chosen parameter values, requiring exponentially many measurement shots to identify a minimizing direction.The paper defines a barren plateau through exponential concentration of loss values with problem size.
- Barren plateaus: For a 100 × 50 circuit, the estimated probability of gradients exceeding ε = 10^-3 is 10^-4, versus 10^-24 for a 100 × 1000 circuit.These estimates use a 1D hardware-efficient ansatz with a local loss and are based on crude numerical extrapolation.
- Training regimes: Avoiding barren plateaus in standard provable cases may yield classically simulable training, while quantum resources could remain useful for data collection, inference, or sampling.The paper also notes a converse possibility in which training requires a quantum computer but testing is classical.
- Barren plateaus: Barren plateaus are an average-case notion, so useful gradient trajectories near suitable initializations may still exist, although finding them in practice remains open.Practitioners need not explore most parameter values, only trajectories toward reasonable solutions.
- Optimization landscapes: Local minima and oscillatory Fourier components complicate variational optimization landscapes.The cost function can contain many local minima where optimizers become stuck.
- Overparametrization: Quantum overparametrization may require deep or wide circuits, potentially creating near-term hardware challenges and additional barren plateaus.Classical overparametrization can turn local minima into saddle points and provide more descent directions, but the quantum implementation is constrained.
- Overparametrization: No-cloning prevents straightforward reuse of intermediate results during quantum backpropagation, while copy complexity and degraded precision scaling remain unresolved obstacles.The paper identifies efficient overparametrization in NISQ-friendly settings as an open problem.
- Open prospects: Problem-inspired models with special initializations may remain trainable, and non-variational quantum subroutines could enhance classical variational methods.The prospects of these approaches remain undetermined rather than ruled out.
2. Variational quantum algorithms beyond NISQ
Variational quantum algorithms are not confined to the NISQ era: they remain relevant for approximations in fault-tolerant computing, although implementation challenges remain. Progress on circuit costs, error mitigation, and amplitude estimation supports their continued development.
- Variational methods remain relevant beyond NISQ because they provide a natural way to find approximations when exact approaches are difficult.
- Fault-tolerant implementations still face open questions, especially high T-gate costs from fine rotation-angle resolution.
- Quantum error mitigation can address circuit-compilation errors from finite rotation resolution and algorithmic errors from finite Trotter approximations.
- Fault tolerance may reduce expectation-value estimation costs through amplitude amplification, while classical shadows may reduce circuit repetitions.
- Despite technical challenges involving repetition counts and rotation-angle resolution, some variational algorithms will likely find useful fault-tolerant applications.
C. Exponential speedups in applications
The absence of proven exponential speedups with guaranteed commercial value is correct, but it does not rule out practical quantum advantages. For fixed-size scientific problems and empirically validated heuristics, effectiveness against classical methods remains the central test.
- No proven exponential quantum speedups currently guarantee substantial commercial value in machine learning, optimization, quantum chemistry, or materials science.
- Shor’s algorithm supports superpolynomial speedups for computer algebra, but commercial applications generally lack comparable mathematical structure.
- For fixed-size quantum-chemistry problems, the relevant question is practical advantage rather than asymptotic complexity.
- Classical heuristics can exploit structure in practical machine-learning and optimization instances, making quantum heuristics difficult to validate without suitable hardware.
- Fault-tolerant machines may enable empirically validated quantum heuristics and potentially large provable speedups for commercially relevant problems.
- Meaningful speedups in practical problems remain reasonable to expect even though proving exponential end-to-end advantages is mathematically challenging.
III. OUTLOOK
Practical quantum advantage is constrained by noise, circuit-training difficulty, and competition from classical algorithms. The outlook favors modest-depth, problem-informed approaches, with late NISQ and early fault-tolerant stages opening progressively broader opportunities.
- Practical quantum advantage must overcome hardware noise, circuit-training hardness, and advanced classical algorithms.
- Promising applications require modest circuit depth, problem-motivated Ansätze, classical precomputation, suitable cost functions and initial states, and favorable application structure.
- The late NISQ era may support earliest practical advantages with a few tens of thousands of gates, especially for many-body dynamics beyond current classical methods.
- Discrete-time evolutions such as Floquet dynamics are highlighted as promising late-NISQ targets, while continuous-time simulation faces Trotterization overhead.
- Early fault-tolerant machines with gate errors of order 10^-6–10^-8 could support deeper circuits and algorithms including phase estimation, quantum signal processing, and Gibbs sampling.
- Quantum heuristics may eventually become as broadly useful as classical heuristic methods, which originated in physics simulation and later spread widely.