Representation-Induced Symmetry Trapping in Adaptive Variational Quantum Simulations of Multi-Reference Topologies

Abstract: Evaluating the trainability of adaptive quantum chemistry algorithms under multi-reference static correlation requires understanding how representation topologies intertwine with molecular geometry. We systematically expose a deep physical dependence on point-group symmetry by evaluating a spin-conserved SUSD operator pool across highly stretched configurations (2 x Re) of asymmetric LiH, symmetric BeH2, and asymmetric H2O. Under asymmetric distortions, the non-local mapping constraints of the Bravyi-Kitaev transformation create an optimization trapping effect--an encodement-locked manifestation of the broader barren plateau crisis. Crucially, by comparing these to the symmetrical stretching baseline of BeH2, we demonstrate that the preservation of point-group symmetry structurally protects the optimization landscape, proving that ansatz symmetry restrictions are necessary but insufficient without accounting for the underlying fermion-to-qubit representation. While current methods rely on numerical pruning to throttle pool sizes, our structural approach establishes that the mapping representation remains a critical factor in maintaining landscape trainability. Furthermore, exploiting structural overlap within our pool, we introduce a covariance-driven, adaptive shot-allocation filter. Diverging from static energy-variance minimization frameworks, our allocation engine operates as a dynamic runtime diagnostic tool. By continuously monitoring the gradient precision threshold epsilon, it aggressively prunes dead symmetry channels and triggers an automated circuit-termination sequence upon detecting representation-induced flat-lined states (dE/dtheta approx 0). This integration of algebraic measurement reuse with topology-aware statistical filtering provides a promising, resource-efficient strategy for executing deep variational algorithms on early fault-tolerant architectures.

• The paper reveals that adaptive variational simulations suffer from representation-induced symmetry trapping, impeding classical optimizer efficiency under BK mapping.
• It deploys a spin-conserved operator pool and dynamic measurement strategies to enforce symmetry protection and enhance resource efficiency.
• Numerical results show instant convergence in symmetric cases, while asymmetric molecules face non-convex optimization and operator ping-pong phenomena.

Representation-Induced Symmetry Trapping in Adaptive Variational Quantum Simulations

Introduction

The paper "Representation-Induced Symmetry Trapping in Adaptive Variational Quantum Simulations of Multi-Reference Topologies" (2606.13387) rigorously investigates the interplay between molecular geometry, point-group symmetry, and fermion-to-qubit mapping representations in the context of adaptive variational quantum chemistry algorithms. It systematically analyzes how these structural factors influence classical optimization trainability, particularly for strongly correlated, multi-reference dissociation regimes that pose substantial challenges to NISQ-era quantum simulation.

Theoretical Framework and Operator Pool Design

The authors deploy a spin-conserved Singlet-Unrestricted Singles and Doubles (SUSD) operator pool, ensuring the variational workflow remains confined to singlet manifolds regardless of structural distortion or mapping representation. Unlike numerically pruned pools, this framework enforces spatial symmetry restrictions algebraically, yielding tightly bounded pool sizes (e.g., 8 for H₂O, 16 for LiH, 204 for BeH₂). Explicit analytical gradients for each pool operator are evaluated under two mapping representations: Jordan-Wigner (JW) and Bravyi-Kitaev (BK).

Crucially, the study shifts focus from conventional metrics favoring BK's reduced Pauli string weight (logarithmic in system size) to the impact of mapping locality on the gradient landscape explored by the classical optimizer. The commutator algebra—core to selection of ansatz layers in ADAPT-VQE—is fundamentally altered by mapping choice, which in turn modulates optimization dynamics in the presence of broken molecular symmetry.

Numerical Results: Convergence Dynamics and Symmetry Trapping

Comprehensive simulations were performed for highly stretched (2×Re) regimes of asymmetric LiH and H₂O versus symmetric BeH₂. The results display a stark mapping-induced divergence:

• Symmetric molecules (BeH₂): Both JW and BK mappings yield mapping-invariant, instant convergence to the ground state (zero circuit expansion, energy attains the reference value).
• Asymmetric molecules (LiH, H₂O): JW produces rapid, clean convergence with minimal circuit depth (4 or 3 operators, full energy relaxation). In contrast, BK triggers severe optimization paralysis—characterized by cyclic operator loops, enormous residual gradients, and zero structural energy relaxation despite linear circuit depth expansion (up to hardware termination limits).

The analysis reveals that the hierarchical parity-tracking structure of BK fractures the classical gradient landscape when spatial symmetry is broken. The adaptive optimizer becomes trapped in a degenerate subspace, exhibiting the "operator ping-pong" phenomenon and failing to reduce active space energy. This encodement-locked behavior is tied to a broader barren plateau crisis in quantum neural network training, but here emerges through explicit geometric and representation interactions.

Quantitatively, the gradient residuals in the stalled BK runs remained massive (e.g., 37.11 for H₂O, 1.77 for LiH), contrasting the instant collapse of symmetry-preserved cases. This underscores that ansatz symmetry restrictions alone are insufficient for landscape trainability unless coupled with mapping locality.

Measurement Overhead and Statistical Shot Optimization

Adaptive quantum chemistry on NISQ hardware faces prohibitive measurement costs: analytical gradients for each pool operator must be resolved at every iteration, scaling as $O(N^4)$. The paper addresses this through:

1. Pauli Measurement Reuse: Structural overlap of commutators within the SUSD pool enables pooling of repeated Pauli strings across the operator set; grouped measurement of TPB cliques reduces distinct circuit configurations by up to two orders of magnitude for large systems.
2. Covariance-Driven Adaptive Shot Allocation: A runtime allocation engine dynamically distributes measurement shots in proportion to variance and gradient relevance, set by a tunable threshold $\epsilon$. This system aggressively prunes dead symmetry channels and rapidly flags stagnant optimization trajectories, especially during representation-induced traps (e.g., flat-lined $dE/d\theta$ under BK mapping). The diagnostic routine halts circuit expansion, safeguarding hardware resources.

This practical innovation diverges from existing static energy-variance minimization frameworks: shot allocation is dynamically modulated by live topology, ensuring efficiency and robustness across varied molecular and mapping landscapes.

Implications and Future Directions

The findings invalidate the assumption of inherent superiority of hierarchical mapping in adaptive variational algorithms. The preservation of point-group symmetry structurally protects the optimization landscape, but this is not universally applicable without careful mapping selection. For asymmetrically distorted systems and multi-reference regimes, BK mapping introduces severe non-convexity and gradient fragmentation, presenting a fundamental trade-off between reduced operator weight and trainability.

Practically, this necessitates algorithmic strategies that account for mapping-induced symmetry traps. The proposed measurement reuse and adaptive shot allocation protocols provide resource-efficient blueprints for executing deep variational circuits on NISQ and early fault-tolerant architectures.

Theoretically, the emergent mapping-invariance and instant convergence in symmetric cases hint at the possibility of algorithmically engineered reference transformations targeting explicit symmetry protection. Formal group-theoretic characterization of trap conditions remains an open avenue.

Conclusion

This study exposes deep structural dependencies linking molecular topology, symmetry, and fermion-to-qubit representation with optimization landscape feasibility in adaptive quantum chemistry. By combining rigorous operator pool design, numerical benchmarking, and dynamic measurement resource management, it sets stringent constraints and practical guidelines for the deployment of quantum simulation algorithms under strong correlation. Future research avenues include rigorous formalization of symmetry trap conditions and further development of topology-aware diagnostic tools to exploit instant convergence regimes and mitigate representation-induced fragmentation.