Expressibility and trainability of a two-dimensional pairwise quantum-circuit ansatz

Ji-Chong Yang; Shuai Zhang; Wei Liu

A native 2D pairwise ansatz is more expressive at shallow layers and has smaller early gradient variance than three 1D ansatze on a fixed 16-qubit system when compared at equal layer counts.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-15 01:41 UTC pith:6SWNOFRI

load-bearing objection Abstract-only 2D pairwise ansatz comparison on 16 qubits: plausible hardware-native baseline, but equal-L rankings are confounded by denser layers and cannot be verified. the 3 major comments →

arxiv 2607.12996 v1 pith:6SWNOFRI submitted 2026-07-14 quant-ph

Expressibility and trainability of a two-dimensional pairwise quantum-circuit ansatz

Shuai Zhang , Wei Liu , Ji-Chong Yang This is my paper

classification quant-ph

keywords parameterized quantum circuitsvariational quantum algorithmsexpressibilitytrainability2D pairwise ansatzframe potentialgradient variancehardware-efficient ansatz

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Parameterized quantum circuits used in variational algorithms and quantum machine learning are often built from simplified 1D chain or ring entanglement patterns, even though real superconducting hardware is typically 2D. This paper constructs a native 2D pairwise ansatz that follows planar connectivity and measures its expressibility and trainability against three representative 1D ansatze. On a fixed 16-qubit system, the comparison is performed at identical layer counts L even though the circuits then have different total depths. At L=1 and L=2 the 2D circuit yields the smallest Kullback–Leibler divergence from the Haar distribution, and its second-order frame potential approaches the theoretical lower bound faster than the 1D circuits at shallow depth. Gradient variance of the full Pauli-Z string expectation with respect to the first Ry parameter is also smaller for the 2D circuit through L=4; the gap narrows thereafter and disappears by L=6. The practical claim is that matching the hardware’s native 2D geometry can improve both expressibility and early trainability without adding layers.

Core claim

For a fixed 16-qubit register the native 2D pairwise ansatz attains the smallest KL divergence at layer counts L=1 and L=2, drives its second-order frame potential toward the Haar lower bound more rapidly at shallow layers than three standard 1D ansatze, and produces a smaller gradient variance of the observable ⟨Z0⊗⋯⊗Z15⟩ with respect to the first Ry angle for L=1–4, all when the circuits are compared at identical layer depth.

What carries the argument

The native 2D pairwise ansatz: a layered parameterized circuit whose two-qubit gates follow the planar nearest-neighbor edges of a 2D grid, used as the central object whose expressibility (KL divergence and second-order frame potential) and trainability (gradient variance of a full Pauli-Z string) are measured against 1D counterparts at equal layer count L.

Load-bearing premise

That equal layer depth L is a fair comparison axis for expressibility and trainability even though the 2D and 1D circuits then possess different total gate depths and therefore different noise exposure.

What would settle it

Recompute the same KL divergences, frame potentials and gradient variances on the identical 16-qubit observables after equalizing total circuit depth (or total two-qubit gate count) rather than layer count L; if the 2D advantage disappears, the claimed superiority at equal L is an artifact of the comparison protocol.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Abstract-only 2D pairwise ansatz comparison on 16 qubits: plausible hardware-native baseline, but equal-L rankings are confounded by denser layers and cannot be verified.

read the letter

This is an abstract-only look at a hardware-motivated 2D pairwise PQC ansatz on 16 qubits, ranked against three 1D baselines on KL expressibility, second-order frame potential, and one global Z-string gradient variance, all at equal layer count L.

What is new is the concrete native 2D pairwise construction and the specific shallow-layer numerical rankings. That is a legitimate, limited-scope extension of the existing ansatz-design literature rather than a new framework. The abstract is clear about the metrics and the system size, and it does not hide the fact that circuit depths differ. Credit for writing a clean, falsifiable comparison protocol and for flagging the depth disparity up front.

The soft spot that actually matters is the comparison axis. A planar 2D pairwise layer has higher average degree (~4 vs ~2), so equal L means more two-qubit gates, more parameters, and different depth. The abstract itself notes the depth mismatch yet still reports rankings at identical L. Without a resource-matched control (gate count, parameter count, or depth), you cannot cleanly attribute the better KL and faster frame-potential approach to 2D geometry rather than denser layers. That is the load-bearing premise of every ranking. Secondary caveats: smaller gradient variance is often a barren-plateau signal, so calling it “trainability” needs careful interpretation in the full text; results are for one n and one observable.

I cannot check methods, error bars, or code from the abstract alone, so confidence stays low. The work is still serious: it engages standard Haar benchmarks and reports concrete numbers rather than circular fitting. It is for people already working on VQA/QML ansatz design for planar hardware who want a 2D baseline and shallow-layer metric numbers. It deserves a serious referee if the full paper supplies the missing resource-matched ablations and data; otherwise the central claim stays under-supported. I would send it to review rather than desk-reject, with the equal-L confound as the main revision request.

Referee Report

3 major / 4 minor

Summary. The manuscript proposes a native 2D pairwise parameterized quantum circuit ansatz motivated by planar superconducting connectivity and compares its expressibility and trainability to three representative 1D ansatze on a fixed 16-qubit system. Expressibility is assessed via KL divergence of the fidelity distribution relative to Haar and via the second-order frame potential; trainability is assessed via the variance of the gradient of ⟨Z0⊗⋯⊗Z15⟩ with respect to the first Ry parameter. At equal layer count L, the abstract reports that the 2D ansatz has the smallest KL at L=1,2, approaches the Haar frame-potential lower bound more rapidly at shallow L, and exhibits smaller gradient variance at L=1–4 (with differences narrowing by L=5 and statistical compatibility at L=6), while noting that the circuits have different actual depths.

Significance. If the reported advantages can be attributed to 2D connectivity rather than to unmatched resource counts, the work would be a useful, hardware-motivated contribution to ansatz design for VQAs and QML on planar devices. The use of standard expressibility diagnostics (KL, frame potential) and a concrete gradient-variance statistic is appropriate and falsifiable in principle. Significance is currently limited by the equal-layer comparison protocol and by the restriction to a single system size, one observable/parameter pair, and abstract-only numerical claims that cannot yet be audited against methods or error bars.

major comments (3)

[Abstract (comparison protocol)] The central rankings (KL at L=1,2; faster approach of the second-order frame potential at shallow L; smaller gradient variance of ⟨Z0⊗⋯⊗Z15⟩ at L=1–4) are all reported at identical layer depth L, while the abstract itself states that the circuits have different circuit depths. A planar 2D pairwise layer has higher average degree (~4 vs ~2) and therefore more two-qubit gates and typically more free parameters per layer than a 1D chain/ring layer. Equal L therefore equates neither gate count, parameter count, nor compiled depth. Without resource-matched controls (e.g., equal two-qubit gate count, equal parameter count, or equal compiled depth), the reported advantages cannot be attributed to 2D geometry rather than denser layers. This premise underpins every ranking in the abstract and is load-bearing for the hardware-inspired claim.
[Abstract (gradient-variance paragraph)] Gradient trainability is reported only for a single global Pauli-Z string and a single parameter (the first Ry angle) on n=16. Barren-plateau and trainability conclusions are known to depend strongly on the observable locality, the parameter location, and system size. The abstract’s trainability ranking is therefore under-supported as a general statement about the 2D ansatz; either a broader observable/parameter scan or a clearly scoped claim limited to this observable is needed.
[Abstract (all numerical claims)] All numerical claims are given for a single system size (16 qubits) and without reported sample sizes, error bars, or statistical tests beyond the qualitative phrase “statistically compatible” at L=6. For KL, frame potential, and gradient variance, the manuscript must specify ensemble sizes, estimators, and uncertainty quantification so that the shallow-L rankings can be audited. With only the abstract available, these results cannot be verified.

minor comments (4)

[Abstract] Define the three “representative 1D ansatze” explicitly (connectivity pattern, gate set, and parameter placement) so that the comparison is reproducible from the text alone.
[Abstract] Clarify what “layer” means for the 2D pairwise ansatz (which edges are covered per layer; whether the layer is a matching, a full planar neighborhood, or a fixed tiling) and how that definition maps to circuit depth.
[Abstract] State the precise Haar lower bound used for the second-order frame potential and the estimator (finite ensemble size) so that “approaches more rapidly” is quantitatively checkable.
[Abstract] The phrase “native 2D pairwise ansatz” should be accompanied by a brief gate-level description or figure reference when the full text is available; the abstract alone leaves the construction underspecified.

Circularity Check

0 steps flagged

No significant circularity: abstract reports numerical ensemble comparisons against external Haar benchmarks, not self-derived predictions.

full rationale

Only the abstract is available. It describes construction of a native 2D pairwise ansatz and numerical comparison of expressibility (KL divergence of the fidelity distribution; second-order frame potential vs. the known Haar lower bound) and trainability (gradient variance of a fixed Pauli-Z string expectation) against three representative 1D ansatze at equal layer count L on 16 qubits. These quantities are standard external benchmarks for PQC ensembles; the reported rankings are empirical outcomes of sampling circuit ensembles, not quantities forced by definition, by a fitted parameter re-labeled as a prediction, or by a load-bearing self-citation chain. No uniqueness theorem, ansatz-smuggling citation, or renaming of a known empirical law appears. Equal-L comparison may confound topology with gate density (a methodological fairness issue), but that is not circularity under the stated criteria. With no self-definitional step, no fitted-input-as-prediction, and no load-bearing self-citation visible, the circularity score is 0 and steps is empty.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 1 invented entities

Abstract-only review: free parameters and axioms are those implied by the stated experimental design. No invented physical entities. The central rankings rest on standard Haar-expressibility diagnostics, a single global observable for gradients, fixed n=16, and the equal-layer comparison convention.

free parameters (3)

system size n = 16
Fixed at 16 qubits for all reported comparisons; not derived, chosen as the experimental setting.
layer depth L = 1–6
Discrete layer counts L=1..6 used as the independent variable; equal-L matching is a design choice, not a derived optimum.
observable and parameter for gradient variance = ⟨Z^⊗16⟩ vs first Ry
Gradient variance is reported only for ⟨Z0⊗⋯⊗Z15⟩ w.r.t. the first Ry angle; choice of this string and parameter is a free experimental selection that defines the trainability probe.

axioms (3)

domain assumption KL divergence from Haar and second-order frame potential are valid proxies for PQC expressibility.
Standard in the expressibility literature; invoked as the basis for ranking the 2D vs 1D ansatze at shallow L.
domain assumption Gradient variance of a single global Pauli-Z string w.r.t. one rotation angle is a meaningful trainability diagnostic.
Used as the trainability metric in the abstract; may not capture full optimization landscape or local observables.
ad hoc to paper Equal layer depth is a fair comparison across topologies with different circuit depths.
Explicitly stated comparison protocol; different gate counts/depths are acknowledged but not equalized.

invented entities (1)

native 2D pairwise ansatz no independent evidence
purpose: Hardware-aligned entanglement pattern using only planar nearest-neighbor pairwise gates as the variational circuit template.
The ansatz structure is the paper’s constructed object of study; independent evidence would require performance on actual VQA tasks or hardware, not only the abstract’s metric rankings.

pith-pipeline@v1.1.0-grok45 · 6153 in / 2833 out tokens · 33459 ms · 2026-07-15T01:41:57.304641+00:00 · methodology

0 comments

read the original abstract

Parameterized quantum circuits~(PQCs) constitute a central building block of variational quantum algorithms~(VQAs) and quantum machine learning~(QML) methods. Existing ansatz designs often adopt hardware-agnostic or simplified 1D chain/ring entanglement patterns. However, as quantum hardware continues to develop, native 2D connectivity patterns, such as planar superconducting-qubit architectures, are becoming increasingly important. Inspired by this hardware structure, we construct a native 2D pairwise ansatz and compare its expressibility and trainability with representative 1D ansatze at identical layer depths, despite their different circuit depths. For the fixed 16-qubit system, the 2D ansatz has the smallest KL divergence at $L=1$ and $2$, and its second-order frame potential approaches the theoretical lower bound more rapidly at shallow layer counts than the frame potentials of the three 1D ansatze. We also evaluate the gradient variance of the Pauli-$Z$-string expectation value $\langle Z_0\otimes\cdots\otimes Z_{15}\rangle$ with respect to the first $R_y$ angle. For this Pauli-$Z$ string and fixed parameter, the gradient variance is smaller for the 2D circuit at $L=1$--$4$. The differences narrow at $L=5$, and the four ansatze yield statistically compatible variances at $L=6$.

Expressibility and trainability of a two-dimensional pairwise quantum-circuit ansatz

Core claim

What carries the argument

Load-bearing premise

What would settle it

discussion (0)