arXiv:2609.20345
Dissipative phase transitions in quantum reservoir computing
Da Zhang, Zhang-Qi Yin
Abstract
Enhanced performance of quantum reservoir computing has been associated with dynamical phase transitions, but whether this connection extends to dissipative systems and which relaxation mechanisms underlie it remain insufficiently understood. We systematically compare driven-dissipative Kerr reservoirs across first-order and continuous dissipative phase transitions and find enhanced memory and nonlinear processing near both phase boundaries. Although the closure of the Liouvillian gap marks the transitions in the thermodynamic limit, computational performance does not generally follow gap suppression. An exact post-training Liouvillian-mode decomposition quantitatively attributes the trained memory to intrinsic relaxation channels. It shows that the gap mode contributes only weakly, while finite-rate modes and their cross-contributions dominate the enhanced capacity. These results go beyond phenomenological correlations by directly linking memory capacity to the intrinsic Liouvillian relaxation spectrum. Moreover, these findings provide a physical basis for designing dissipative quantum reservoirs and testing memory-enhancement mechanisms in experimentally accessible Kerr platforms.
AI Summary
Does improved quantum-reservoir performance near dissipative phase transitions arise from critical slowing of the Liouvillian gap, or from other relaxation channels? This paper numerically compares two single-mode driven-dissipative Kerr reservoirs: a coherently driven oscillator with a first-order transition and a two-photon-driven oscillator with a continuous $Z_2$-symmetry-breaking transition. Inputs modulate a quadrature drive, six observables are sampled at virtual times, and only an affine readout is trained. Performance is assessed through delayed Legendre-polynomial reconstruction, using degrees $1$–$3$ and delays $0$–$9$, and NARMA prediction tasks. Both reservoirs perform best near their transition regions. At the first-order transition, this coincides with compression of the low-lying relaxation spectrum. For the continuous transition, performance declines deeper in the symmetry-broken regime even as the gap continues to soften. An exact post-training decomposition expands the driven trajectories and readout in the zero-input Liouvillian eigenbasis, attributing the measured capacity to decay-rate bands and signed cross-band terms. The gap-containing slow sector contributes weakly; finite-rate sectors dominate. Near first-order coexistence, reduced destructive cross-band cancellation accompanies the capacity maximum, whereas the continuous-transition optimum involves increased positive within-band contributions. The attribution is task- and protocol-dependent, and band contributions are signed allocations, not independently achievable capacities. The spectrum alone cannot predict performance because input excitation, measurement visibility, and readout weights also matter. Experimental access through relaxation or correlation spectroscopy is proposed, not demonstrated; only modes with nonzero preparation and measurement overlaps are visible. Extension is discussed for other diagonalizable Markovian reservoirs.
Central claim may not hold: The abstract's claim that the individual Liouvillian gap mode contributes weakly is not established by the reported weak signed attribution of an entire slow-mode band, which can conceal cancellation between its constituent modes.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 8 pages, 4 figures