arXiv:2609.30496
Disentangling Expressibility, Symmetry Protection, and Hardware Noise in Variational Quantum Simulation of the Two-Flavor Schwinger Model
Karthikeya Machiraju, Krishna Sujith, Kaustav Bhowmick
Abstract
Existing quantum simulations of the two-flavor Schwinger model have run at a single lattice size, and it is not known how far the variational approach can be pushed or which weakness stops it first. Following the model from N = 2 to 6 staggered lattice sites, we find that the binding constraint at reachable sizes is hardware noise rather than circuit expressibility or trainability, and identify N = 3 as the immediately viable extension of existing trapped-ion experiments. The energy error of a charge-conserving ansatz collapses onto one function of p/d, the ratio of variational parameters to physical-sector dimension, and falls by more than two orders of magnitude as p/d rises through order unity, giving the expressibility condition L(4N - 1) >= binom(2N,N) for L circuit layers. The condition is local in chemical potential: at N = 3 the layer count sufficient at zero chemical potential leaves a 74.38% error near the first-order boundary, while one further layer reaches 0.08%. Charge conservation also protects trainability and prevents charge-sector leakage: as the qubit count doubles from 4 to 8, the normalized gradient variance falls to 1/3.56 of its starting value for the constrained ansatz, versus 1/13.57 for an unconstrained circuit. Comparing a global contraction with per-gate local noise, a fixed-parameter control shows that the noise model, not whether the optimizer runs inside the noisy loop, sets how strongly noise degrades the first-order transition. At N = 4, a noiseless control reaches 0.12% mean error, whereas the same circuit at 1.00% depolarizing noise reaches 25.69 to 52.81%.
AI Summary
The paper asks which limitation—expressibility, trainability, or hardware noise—first obstructs VQE for the charge-neutral, two-flavor Schwinger model. It studies $N=2$–$6$ staggered sites (4–12 qubits) at zero bare mass and coupling $x=16$, using exact diagonalization, statevector optimization, density-matrix noise simulations, and a six-qubit hardware evaluation. For a layered charge-conserving exchange ansatz with $p=L(4N-1)$ parameters and sector dimension $d=\binom{2N}{N}$, energy errors approximately collapse against $p/d$, motivating the heuristic depth criterion $p\gtrsim d$. Crossings are tested only at $N=2,3,4$; gate ordering produces an exception. The criterion is also chemical-potential dependent: at $N=3$, two layers give 74.38% error near a first-order boundary, while three reduce it to 0.08%. Charge conservation prevents leakage into lower-energy nonphysical sectors and gives slower normalized gradient-variance decay: a factor 3.56 versus 13.57 for an unconstrained circuit between four and eight qubits. Fixed-parameter controls indicate that local per-gate noise, rather than noisy-loop reoptimization, explains weaker transition-slope suppression than global depolarizing contraction: approximately 10–16% versus 26.77% at $N=3$. At $N=4$, a five-layer, 70-CNOT circuit has 0.12% mean noiseless error but 25.69–52.81% error under 1% per-CNOT depolarization, identifying noise as limiting in that setting. Stated limitations include incoherent-only simulated noise, COBYLA optimization, restricted ansatz/connectivity comparisons, untested threshold crossings at $N=5,6$, and a retained-signal model that can yield values above unity. Generality across couplings, boundary conditions, gate orderings, and discretizations remains untested.
Central claim may not hold: The expressibility crossover is tested only at N=2,3,4, while the N=5,6 calculations remain far below the proposed threshold and cannot establish its larger-size scaling. The claimed single-function collapse has an explicit exception: Table vqe reports 49.26% error at p/d=1.167 and 0.01% at p/d=1.10. The proposed necessary expressibility condition does not follow from parameter counting, because representing one particular ground state does not require covering the entire physical-state manifold. The nearest-neighbor XY-plus-Rz ansatz generates only number-conserving fermionic Gaussian operations, so adding layers cannot make it complete in the exponentially large neutral sector as the Discussion claims.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.