Quantum natural gradient has emerged as a superior minimisation technique in quantum variational algorithms. Classically simulating the algorithm running on near-future quantum hardware is paramount in its study, as it is for all variational algorithms. In this case, state-vector simulation of the P-parameter/gate ansatz circuit does not dominate the runtime; instead, calculation of the Fisher information matrix becomes the bottleneck, involving O(P^3) gate evaluations, though this can be reduced to O(P^2) gates by using O(P) temporary state-vectors. This is similar to the gradient calculation subroutine dominating the simulation of quantum gradient descent, which has attracted HPC strategies and bespoke simulation algorithms with asymptotic speedups. We here present a novel simulation strategy to precisely calculate the quantum natural gradient in O(P^2) gates and O(1) state-vectors. While more complicated, our strategy is in the same spirit as that presented for gradients in Reference 6, and involves iteratively evaluating recurrent forms of the Fisher information matrix. Our strategy uses only "apply gate", "clone state" and "inner product" operations which are present in practically all quantum computing simulators. It is furthermore compatible with parallelisation schemes, like hardware acceleration and distribution. Since our scheme leverages a form of the Fisher information matrix for strictly unitary ansatz circuits, it cannot be simply extended to density matrix simulation of quantum natural gradient with non-unitary circuits.
Quantum natural gradient has emerged as a superior minimisation technique in quantum variational algorithms. Classically simulating the algorithm running on near-future quantum hardware is paramount in its study, as it is for all variational algorithms. In this case, state-vector simulation of the P-parameter/gate ansatz circuit does not dominate the runtime; instead, calculation of the Fisher information matrix becomes the bottleneck, involving O(P^3) gate evaluations, though this can be reduced to O(P^2) gates by using O(P) temporary state-vectors. This is similar to the gradient calculation subroutine dominating the simulation of quantum gradient descent, which has attracted HPC strategies and bespoke simulation algorithms with asymptotic speedups. We here present a novel simulation strategy to precisely calculate the quantum natural gradient in O(P^2) gates and O(1) state-vectors. While more complicated, our strategy is in the same spirit as that presented for gradients in Reference 6, and involves iteratively evaluating recurrent forms of the Fisher information matrix. Our strategy uses only "apply gate", "clone state" and "inner product" operations which are present in practically all quantum computing simulators. It is furthermore compatible with parallelisation schemes, like hardware acceleration and distribution. Since our scheme leverages a form of the Fisher information matrix for strictly unitary ansatz circuits, it cannot be simply extended to density matrix simulation of quantum natural gradient with non-unitary circuits.
Calculating the energy gradient in parameter space has become an almost ubiquitous subroutine of variational near-term quantum algorithms. "Faithful" classical emulation of this subroutine mimics its quantum evaluation, and scales as O(P^2) gate operations for P variational parameters. This is often the bottleneck for the moderately-sized simulations, and has attracted HPC strategies like "batch-circuit" evaluation. We here present a novel derivation of an emulation strategy to precisely calculate the gradient in O(P) time and using O(1) state-vectors, compatible with "full-state" state-vector simulators. The prescribed algorithm resembles the optimised technique for automatic differentiation of reversible cost functions, often used in classical machine learning, and first employed in quantum simulators like Yao.jl. In contrast, our scheme derives directly from a recurrent form of quantum operators, and may be more familiar to a quantum computing community. Our strategy is very simple, uses only 'apply gate', 'clone state' and 'inner product' primitives and is hence straightforward to implement and integrate with existing simulators. It is compatible with gate parallelisation schemes, and hardware accelerated and distributed simulators. We describe the scheme in an instructive way, including details of how common gate derivatives can be performed, to clearly guide implementation in existing quantum simulators. We furthermore demonstrate the scheme by implementing it in Qiskit, and perform some comparative benchmarking with faithful simulation. Finally, we remark upon the difficulty of extending the scheme to density-matrix simulation of noisy channels.
Variational quantum circuits are promising tools whose efficacy depends on their optimisation method. For noise-free unitary circuits, the quantum generalisation of natural gradient descent was recently introduced. The method can be shown to be equivalent to imaginary time evolution, and is highly effective due to a metric tensor reconciling the classical parameter space to the device's Hilbert space. Here we generalise quantum natural gradient to consider arbitrary quantum states (both mixed and pure) via completely positive maps; thus our circuits can incorporate both imperfect unitary gates and fundamentally non-unitary operations such as measurements. Whereas the unitary variant relates to classical Fisher information, here we find that quantum Fisher information defines the core metric in the space of density operators. Numerical simulations indicate that our approach can outperform other variational techniques when circuit noise is present. We finally assess the practical feasibility of our implementation and argue that its scalability is only limited by the number and quality of imperfect gates and not by the number of qubits.
Parameterized quantum circuits (PQCs) are central to variational quantum algorithms (VQAs), yet their performance is hindered by complex loss landscapes that make their trainability challenging. Quantum natural gradient descent, which leverages the geometry of the parameterized space through quantum generalizations of the Fisher information matrix, offers a promising solution but has been largely limited to pure-state scenarios, with only approximate methods available for mixed-state settings. This paper addresses this question, originally posed in [Stokes et al., Quantum 4, 269 (2020)], by providing exact methods to compute three quantum generalizations of the Fisher information matrix-the Fisher-Bures, Wigner-Yanase, and Kubo-Mori information matrices-for PQCs initialized with thermal states. We prove that these matrix elements can be estimated using quantum algorithms combining the Hadamard test, classical random sampling, and Hamiltonian simulation. By broadening the set of quantum generalizations of Fisher information and realizing their unbiased estimation, our results enable the implementation of quantum natural gradient descent algorithms for mixed-state PQCs, thereby enhancing the flexibility of optimization when using VQAs. Another immediate consequence of our findings is to establish fundamental limitations on the ability to estimate the parameters of a state generated by an unknown PQC, when given sample access to such a state.
Abstract Variational Quantum Algorithms (VQAs) are widely regarded as the most promising route to practical quantum advantage on Noisy Intermediate-Scale Quantum (NISQ) hardware, but their trainability is threatened by the barren plateau phenomenon, in which cost-function gradients vanish exponentially with system size and circuit depth. Standard first-order optimizers, which treat the parameter space of a parameterized quantum circuit (PQC) as flat Euclidean space, are especially vulnerable to this effect and to the additional gradient suppression introduced by hardware noise. This extended treatment presents (i) a from-first-principles derivation of Quantum Natural Gradient (QNG) descent, tracing its lineage from Amari's classical information geometry [1] through the Fubini-Study metric to the Quantum Fisher Information Matrix (QFIM); (ii) a complete taxonomy of barren-plateau mechanisms (expressibility-induced, entanglement-induced, cost-function-locality-induced, and noise-induced); (iii) a comparative survey of the broader NISQ optimizer landscape (SPSA, Adam, COBYLA, Rotosolve, Quantum Natural SPSA, and reduced-cost QFIM estimators); and (iv) a controlled simulation study comparing QNG against vanilla gradient descent on a 4-qubit weighted MaxCut problem under three physically motivated NISQ noise models. Across 50 independent trials, QNG raised the convergence success rate from 30% to 95%, reduced the mean iteration count to convergence by a factor of roughly 6, and achieved a 16% net wall-clock speedup despite the added per-iteration cost of QFIM estimation. We give the full theoretical basis for these gains, an exhaustive discussion of their limitations, an A-to-Z glossary of the field's terminology, and a staged roadmap toward physical hardware validation. Figure placements throughout the document are annotated with detailed image-generation prompts so that the accompanying visual assets can be produced independently of this text.
Variational quantum algorithms are promising tools whose efficacy depends on their optimization method. For noise-free unitary circuits, the quantum generalization of natural gradient descent has been introduced and shown to be equivalent to imaginary time evolution: the approach is effective due to a metric tensor reconciling the classical parameter space to the device's Hilbert space. Here we generalize quantum natural gradient to consider arbitrary quantum states (both mixed and pure) via completely positive maps; thus our circuits can incorporate both imperfect unitary gates and fundamentally nonunitary operations such as measurements. We employ the quantum Fisher information (QFI) as the core metric in the space of density operators. A modification of the error suppression by derangements (ESD) and virtual distillation (VD) techniques enables an accurate and experimentally efficient approximation of the QFI via the Hilbert-Schmidt metric tensor using prior results on the dominant eigenvector of noisy quantum states. Our rigorous proof also establishes the fundamental observation that the geometry of typical noisy quantum states is (approximately) identical in either the Hilbert-Schmidt metric or as characterized by the QFI. In numerical simulations of noisy quantum circuits we demonstrate the practicality of our approach and confirm it can significantly outperform other variational techniques.
The natural gradient descent optimisation technique is an efficient optimising protocol for broad classes of classical and quantum systems that takes the underlying geometry of the parameter manifold into account by means of using either the Fisher information metric of the classical probability distribution function or the Fubini-Study tensor of the associated parametrised quantum states in the consequent update rules. Even though the natural gradient descent procedure utilises the geometry of the space of probability or states, it is, however, insensitive to the measure of parametrised distance on the space of possible outcomes when the corresponding optimising problem is considered for the expectation value of a classical or quantum observable with respect to the probability distribution or the quantum state. In this work, we introduce a generic optimising principle, where the intrinsic geometry of the space of outcomes has been taken into account suitably, either by using an ambient space construction with a base statistical manifold with the usual Fisher information metric (or the Fubini-Study tensor), where the loss hypersurface is embedded to, or by means of a first-principle construction from the overlap of nearby quantum states on the projective Hilbert space. This construction as well as a family of conformal variants yields a form of loss-aware natural gradient updates that rescale the effective step size while preserving the descent direction. We benchmark the resulting optimisers on variational quantum circuit examples and on a classical neural network task, finding that, while the standard natural gradient remains the most robust on average, the proposed conformal schemes can improve best-case convergence in favourable regimes.