acceptodds
Withdrawn from ICLR 2027

Geometric Saturation Diagnosis: Identifying the Capacity Ceiling of Parameterized Quantum Circuits via QGT Rank

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Choosing circuit depth for parameterized quantum circuits remains largely heuristic: circuits are typically deepened until optimization or hardware constraints bind, with little guidance on when additional layers cease to expand expressive capacity. We introduce *Geometric Saturation Diagnosis* (GSD), a training-free diagnostic that tracks the rank of the quantum geometric tensor (QGT) across depths to identify the saturation depth —the point at which the state manifold stops growing. The QGT rank equals the local tangent-space dimension of the parameterized state manifold; its plateau signals that the circuit has reached its geometric expressive ceiling. GSD further enables layer-wise capacity profiling via per-layer rank increments and parameter-direction importance ranking via the QGT singular value spectrum. Across hardware-efficient and Hamiltonian-inspired ans\"atze (–, TFIM and Heisenberg models), saturation depth satisfies the parameter-budget lower bound with decreasing overhead, matches the clean energy of much deeper circuits with substantially fewer parameters, and noise accumulation grows monotonically beyond saturation—identifying as a practical operating point that balances geometric capacity against noise. These results establish QGT rank growth as a principled, geometry-based diagnostic for circuit architecture selection.

Primary area:
optimization
Keywords:
Variational quantum circuits, Quantum geometric tensor, Expressibility

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_QGezOpenReview ↗
Rating
1
Confidence
2
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The paper studies how to choose the depth of a parameterized quantum circuit without training. Its proposal, Geometric Saturation Diagnosis, follows the rank of the quantum geometric tensor as layers are added and looks for the depth where that rank stops rising. The authors also give a parameter-counting lower bound and a Lie-orbit ceiling for the rank. Experiments use HEA and HVA circuits on TFIM and Heisenberg models with 4–10 qubits, and compare VQE energy and noise robustness at the detected depth with shallower and deeper circuits.

Critical strengths

1.A relevant and well-motivated problem. Picking the depth of a variational circuit is still mostly a matter of trial and error, and on NISQ hardware every extra layer adds noise. A cheap, training-free test for the point where more layers stop opening new directions would therefore be useful in practice. Thinking in terms of QGT rank and tangent space is a natural way to frame that question, and the link to Fisher/NTK redundancy diagnostics in deep learning is reasonable. The scan costs reported in Section 5 (0.20–0.74× one training run for N = 4–8) also suggest the diagnostic is affordable. That is why I think the topic is worth pursuing.

2.Separating geometric capacity from task-relevant capacity. The task-relevant tangent gain τL and Proposition 4 make a point that a plain rank count misses: a new geometric direction does not have to lower the energy to first order. The symmetry-constrained experiment in Table 10 is also a useful illustration that the rank diagnostic can pick up a lower orbit ceiling, which goes beyond HEA.

Critical weaknesses

Overall, the paper shows QGT rank saturation but not that the resulting depth is useful; this decides my recommendation.

  1. The experiments do not show that L* is a better depth choice than alternatives, which is the main practical claim.

  2. No meaningful baselines or compute-matched comparison: GSD is not compared with effective dimension, Larocca et al. (2023), expressibility, ADAPT-style growth, or simple shallow-depth stopping, and the GSD scan cost is excluded.

  3. The recommended depth conflicts with the results: at N=8, L* has 2.3% clean-energy error vs 0.4% at L_max; L_half=10 is comparable clean and better under noise (67.6% vs 90.9% degradation at p=0.01); Fig. 4 favors L_max; Table 9 contradicts the monotone trend.

  4. Training is unconverged: 100 Adam steps at lr=0.05 from random starts, 3–10 seeds, ~2% error; depth comparisons mix expressivity with trainability, and convergence curves, transfer, restarts, and gradient variance are missing.

  5. The noise test mostly measures circuit size: equal p per gate means deeper circuits have more gates; fixed-gate-count, depth-independent error, or shot-noise models are needed.

  6. Scale and inconsistencies limit evidence: saturation only up to N=8; at N=10 no saturation within L_max=12 while the parameter bound needs 69 layers; only two spin Hamiltonians, exact statevector, no shot noise; inconsistent L* values (N=6: 9 vs 8; N=8: 20 vs 21; Fig. 4 uses 20; Table 5 gives L*=5 with L_exact=4).

  7. Theoretical claims overreach: the reported ranks (30, 126, 510) are real tangent-space dimensions, so the paper must clarify whether it ranks Q, Re(Q), or a realified Jacobian; Omega(2^N/N) is only a counting bound, alpha_N->1 is extrapolated from three points, and Eq. (6)/L* as an upper bound are unproven. The 5% L* is not L_exact (rank 454 to 510 between L*=21 and L_exact=25), so “beyond L* no new directions exist” is false as written.

  8. Novelty is unclear relative to Haug, Bharti & Kim (2021) and Larocca et al. (2023), which already use QFI/QGT-rank saturation, effective dimension, and parameter redundancy.

Questions for authors
  1. Could you compare GSD with at least two depth-selection baselines at matched total compute, including the cost of the GSD scan? Suitable baselines include effective quantum dimension or the Larocca et al. (2023) threshold, expressibility-based selection, ADAPT-style growth, or training a few shallow depths with early stopping. Is there a setting where the GSD-selected depth outperforms all of them?

  2. Given Table 2 (2.3% clean error at L* versus 0.4% at L_max, and L=10 better at every nonzero noise level) and Figure 4, what accuracy or cost criterion identifies L* as the operating point?

  3. Do the depth comparisons survive longer optimization, parameter transfer, or restarts? Convergence behavior and gradient-variance measurements, especially for deeper circuits, would help clarify whether saturation reflects expressivity rather than trainability.

  4. Is d_L the rank of the complex Q, of Re(Q), or of a realified Jacobian? Please give the exact definition used in the code and explain how rank 126 at N=6 is consistent with Eq. (1), since the complex rank cannot exceed 2^N−1=63.

  5. Is Eq. (6) a theorem or a heuristic? Given Table 2 and Larocca et al. (2023), what supports the claim that L* is an upper bound on useful depth? What exactly does the paper add beyond Haug et al. (2021) and Larocca et al. (2023), which already describe QFI/QGT-rank saturation and overparameterization?

Reviewer_dP4DOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
3
Soundness
1
Read review
Paper summary

This work proposed Geometric Saturation Diagnosis (GSD) which is a framework for inferring the depth at which PQCs reach their capacity ceiling. The idea is to characterize, through the rank of the quantum geometric tensor, the inherent dimensions of reachable quantum state manifold. With this central idea, the authors introduce a criteria for identifying the depth at which the performance saturates. The work consists of theoretical results relating how QGT rank transforms to reachable-set dimension and parameter configurations. The authors perform experiments on hardware-efficient and Hamiltonian-inspired ansatz with n=4 to 10 qubits. The authors report that saturation of QGT rank is consistent with diminishing improvements in VQE performance.

Critical strengths

Clear research problem Finding the appropriate circuit depth is an important problem in variational quantum computing considering the increasing computational and hardware costs with deeper circuits. In this work, it targts finding the minimal circuit depth from a geometric perspective.

Clear interpretation of the method The framework examines circuit capacity through reachable state manifold dimensionality. The authors use layer-wise rank increments to provide an interpretation of how additional layers can contribute true additional, independent parameter directions.

Good experimental settings The experiments consider various circuit architectures, entanglement patterns, system sizes, and Hamiltonians. The authors provide study on sensitivity to numerical thresholds, the QGT singular-value spectrum, parameter efficiency, and depolarizing noise. The results cover complementary perspectives on the argument.

Critical weaknesses

These are the significant concerns that I have:

1. Inconsistency between the QGT rank definition and the reported manifold dimensions.

I have a significant concern regarding identification of the complex QGT rank with the real reachable quantum-state manifold dimension. For example, Eq. [1] defines the QGT as:

Qij=⟨∂iψ∣∂jψ⟩−⟨∂iψ∣ψ⟩⟨ψ∣∂jψ⟩.Q_{ij}=\langle\partial_i\psi|\partial_j\psi\rangle-\langle\partial_i\psi|\psi\rangle\langle\psi|\partial_j\psi\rangle.

For an NN-qubit pure state, the Hermitian matrix rank is bounded by 2N−12^N-1. Tab. 1 reports saturated ranks of 30, 126, and 510 for N=4,6,8N=4,6,8 corresponding to the real dimension 2N+1−22^{N+1}-2 of the projective Hilbert space. These values are consistent with the maximum rank of the real Fubini–Study metric Re⁡(Q)\operatorname{Re}(Q) but not the complex QGT defined in Eq. (1). Therfore, it requires clarifications as for which matrix is used in the experiments and ensure all used math notations are consistent.

2. I can not distinguish novelty relative to established QFIM-based capacity analysis.

Using quantum geometric information to characterize effective parameter dimension of PQCs have been extensively investigated. In particular,

[1] Haug et al. (2021), Capacity and Quantum Geometry of Parametrized Quantum Circuits [2] Larocca et al. (2023), Theory of Overparametrization in Quantum Neural Networks,

These works have examined the relationship between QFIM rank, effective parameter dimension and circuit archietcture. To my understanding, the underlying thoughts of these works and this work is quite similar and it seems that the manuscript does not sufficiently distinguish its contributions from these established results. Specifically, the connection between metric rank saturation and parameter redundancy is already explored and well-known. Therefore, it seems that the primary addition of this work is the use of thresholding to identify the candidate circuit depth.

Questions for authors

See weakness.

Reviewer_6AkPOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

This paper introduces a method to predict the saturation depth of the ansatz circuit based on the QGT rank. Two methods, threshold- and knee detection-based methods are evaluated and are close to the exact depth. To explain the gap between the optimal depth for VQE, the authors introduce a target-relevant tangent to enhance their framework.

Critical strengths

None

Critical weaknesses
  1. The methodogy section is very hard to read. I failed to get the newly proposed technique from the mixture of previous ones
  2. The evaluation does not compare GSD with existing works on saturation depth prediction. The results on VQE performance even disagree with the claim because target-relavant tangent gain (which is derived on Hamiltonian and is not training-free) performs better.
Questions for authors
  1. How does this work compare with Nagano+,26: https://arxiv.org/pdf/2602.03291
  2. Have Knee-detection, threshold detection, and QGT rank been used for saturation depth prediction in existing works? If so, what is new in this paper?
  3. What is the conclusion of section 4.2?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.