Frustration in spin rings limits quantum algorithm performance

Soumyadip Sahoo and Sahinur Reja demonstrate that increasing the depth of a variational quantum eigensolver (VQE) does not always improve performance, a finding that challenges a core expectation of quantum algorithm design. Their work on a frustrated spin system reveals a “double-descent-like shape” where performance initially degrades at intermediate circuit depths before recovering, specifically at a depth of approximately nine layers for a 10-site ring.

The researchers found that even with optimization techniques like “warm start,” the algorithm stalls at a fidelity of roughly 0.50-0.58 for certain frustration values, indicating that achieving low energy alone is insufficient to guarantee an accurate quantum solution.

Symmetry-Preserving VQE for the J 1, J 2 Heisenberg Ring

For a ten-site spin ring, a variational quantum eigensolver (VQE) using a symmetry-preserving ansatz experienced a degradation in performance around a circuit depth of approximately nine layers before eventually recovering, a pattern the authors describe as a “double-descent-like shape”. This unexpected dip challenges the common assumption that increasing circuit depth invariably improves the accuracy of VQE algorithms, and suggests a more complex relationship between depth and optimization success.

The study focused on the frustrated spin-1/2 J1–J2 Heisenberg ring, a model system in condensed matter physics, and employed SingleExcitationMinus gates initiated from the Néel state. The observed abrupt changes in fidelity, energy error, half-chain entropy, and total spin occurred despite the circuit maintaining a finite gradient, ruling out a barren plateau as the cause of the performance collapse.

The researchers discovered that optimization using a “warm start” technique, which extends a shallow, optimized circuit with additional layers, also encounters limitations at specific frustration values. While effective at improving energy error and fidelity for most J2/J1 ratios, the warm start method stalled at a fidelity of approximately 0.50-0.58 when the ratio of J2/J1 exceeded 0.60 for a twelve-site ring, and at a value of 0.60 for a ten-site ring.

“Removing the non-monotonic depth dependence is therefore not the same as reaching the ground state,” the authors write, emphasizing the need to satisfy fidelity, entropy and total spin simultaneously. The team’s work focused on systems with ten and twelve sites, examining frustration values ranging from J2/J1 = 0.01 to 1.00, and used periodic boundary conditions while preserving total magnetization.

They note that the Hamiltonian also commutes with total spin, labeling eigenstates by (S,S^z), and focused on the Sz = 0 sector, which has a dimension of 252 for N = 10 and 924 for N = 12. The study’s conclusions are currently limited to these specific system sizes and frustration values, but the findings raise important questions about the robustness of VQE algorithms and the challenges of accurately assessing their performance on complex quantum systems. The abrupt changes observed are an optimization failure, not a loss of expressibility or a barren plateau, according to the research.

Depth-Dependent Fidelity Collapse on Frustrated Spin Systems

Performance of variational quantum eigensolvers (VQE) on frustrated spin systems can degrade even as circuit depth increases, a counterintuitive finding revealed by recent simulations. This occurs despite achieving a small energy error and maintaining the total spin in the S = 0 sector, highlighting a critical limitation of the optimization approach. The researchers tested the random-seed behavior of the system with 15 seeds, and the figures show one randomly selected seed.

Instead, fidelity, entropy, and total spin must all be satisfied concurrently. For the twelve-site ring, the Sz = 0 sector has a dimension of 924, and the ansatz has P(L) = 24L parameters. The abrupt changes observed in both ten and twelve-site systems accompany changes in half-chain entropy and total spin, suggesting a complex interplay between these parameters during optimization.

Optimization Failure Distinguishes Collapse from Barren Plateaus

Analysis of frustrated spin systems demonstrates that optimization can stall even when energy minimization appears successful, a finding that refines understanding of the challenges in building practical quantum algorithms. The study, focusing on the J₁-J₂ Heisenberg ring with ten and twelve sites, identifies a specific point where optimization techniques reach their limits.

Initial improvements in fidelity and energy error are not sustained indefinitely. The researchers observed a “double-descent-like shape” in performance with random initialization, where degradation follows initial gains before a potential recovery. This points to a genuine optimization failure, where the algorithm gets stuck in a suboptimal solution despite the capacity to represent the true ground state.

A chained warm start stalls at a fidelity of ≈ 0.50-0.58 near J₂/J₁ = 0.60 (for the ten-site ring) and over J₂/J₁ = 0.60-1.00 (for the twelve-site ring), although the energy error is small and the total spin stays in the S = 0 sector. The researchers found that achieving an accurate result requires concurrent satisfaction of fidelity, entropy and total spin. The study’s findings emphasize the need for a more thorough assessment of quantum algorithm performance than simply minimizing energy.

System Size Impacts Depth Dependence and Fidelity Recovery

Performance degradation at intermediate circuit depths, followed by recovery, occurs at different depths depending on system size. For a ten-site ring, an abrupt change in both energy error and fidelity appears around a depth of L≈9, while twelve-site rings exhibit this behavior between layers 10 and 12. This suggests the location of this performance shift is not fixed by parameter count, but is instead influenced by the number of sites within the system.

The recovery following this initial degradation is not uniform across frustration values. The twelve-site system shows a slower recovery at a depth of L = 16, with energy errors ranging from approximately 6-8 and fidelity only reaching 0.14-0.52.

👉 More information
🗞 Depth-Dependent Collapse and Warm-Start Limits of a Symmetry-Preserving VQE for the J$_1$–J$_2$ Heisenberg Ring
✍️ Soumyadip Sahoo and Sahinur Reja
🧠 ArXiv: https://arxiv.org/abs/2611.22810v1

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Ivy Delaney

Ivy Delaney

Ivy Delaney has been working with neural networks and machine learning since the mid-nineties, back when a couple of hidden layers and a long afternoon of training counted as ambitious. She has watched the field go from academic curiosity to the thing quietly running underneath everything, and she brings that long view to quantum computing. For Quantum Zeitgeist she covers the ground where the two fields meet. That means quantum machine learning and the variational algorithms it leans on, and it also means the less glamorous but more interesting story of classical machine learning already doing real work inside quantum machines, decoding error-correcting codes, calibrating noisy hardware and learning the error models that simulators depend on. She writes about the hardware those algorithms have to run on too, and about the post-quantum cryptography scramble that the same hardware has set off. Her stories typically start with the paper, whether that is peer-reviewed work, conference proceedings or an arXiv preprint, with the source linked so you can hold a claim up against the research it came from. She is unimpressed by benchmarks that will not say what they beat, and by demonstrations that only work in the press release.

Latest Posts by Ivy Delaney: