Abstract
We study whether neural quantum states based on multi-layer feed-forward networks can find ground states which exhibit volume-law entanglement entropy. As a testbed, we employ the paradigmatic Sachdev-Ye-Kitaev model. We find that both shallow and deep feed-forward networks require an exponential number of parameters in order to represent the ground state of this model. This demonstrates that sufficiently complicated quantum states, although being physical solutions to relevant models and not pathological cases, can still be difficult to learn to the point of intractability at larger system sizes. This highlights the importance of further investigations into the physical properties of quantum states amenable to an efficient neural representation.
Introduction
Simulating quantum many-body systems is fundamentally limited by the exponential growth of the Hilbert space with system size. Neural Quantum States (NQS) have emerged as a promising variational ansatz to circumvent this "curse of dimensionality." While NQS have shown success in representing "area-law" states—where entanglement scales with the boundary—their performance on "volume-law" states remains a critical open question. This study investigates whether standard feed-forward neural networks can efficiently represent the ground state of the Sachdev-Ye-Kitaev (SYK) model, a system characterized by maximal chaos and volume-law entanglement.
Models and Methods
The primary testbed is the Sachdev-Ye-Kitaev (SYK) model, which describes fermions with random, all-to-all interactions. This model lacks spatial structure, making it a rigorous test for the representational power of NQS.
The researchers employed Fully-Connected Feed-Forward Neural Networks (FFNNs), exploring both shallow (2-layer) and deep architectures. Using Supervised Learning (SL) on ground states obtained via Exact Diagonalization (ED) for system sizes up to L = 18, they optimized the network parameters to minimize energy error. A 1D Heisenberg model served as a baseline to contrast the learning behavior of low-entanglement states.
Results: The Exponential Wall
The central finding is that NQS fail to provide a compressed representation of the SYK ground state. For the SYK model, the number of parameters ($N_{par}$) required to reach a target energy accuracy of $10^{-3}$ scales exponentially with the system size L.
As shown in the scaling analysis, the network effectively "memorizes" the wave function rather than learning a compressed physical rule. The parameter count required for SYK closely tracks the total dimension of the Hilbert space (dim H). In stark contrast, the same architecture represents the 1D Heisenberg model ground state with a number of parameters that grows only polynomially ($L^2$). This suggests that the difficulty is intrinsic to the volume-law entanglement and lack of spatial structure in the SYK model.
Conclusion
This work demonstrates that NQS are not a universal "silver bullet" for highly entangled quantum states. For systems like the SYK model, general-purpose feed-forward architectures succumb to the same exponential scaling as traditional methods. While depth does not alleviate this issue, future research into specialized architectures—such as those incorporating fermionic symmetries or attention mechanisms—may be necessary to find efficient representations for volume-law physics.
Summary updated