Quantum Many-Body Scars (QMBSs) represent a fascinating phenomenon where rare non-thermal eigenstates exist within otherwise chaotic quantum systems. These states weakly violate the Eigenstate Thermalization Hypothesis (ETH), which generally predicts that quantum systems will thermalize. Unlike typical chaotic eigenstates that follow volume-law entanglement scaling, QMBSs exhibit sub-volume-law entanglement entropy, making them distinct markers of non-thermal behavior.
This research employs quantum convolutional neural networks (QCNNs) to identify both known QMBSs and previously undiscovered non-thermal states. QCNNs, already proven effective in classifying quantum phases of matter, are particularly well-suited for this challenge despite the scarcity of training data, as QMBSs constitute only a tiny fraction of all eigenstates.
The study investigates three models known to host QMBSs: the xorX model, PXP model, and far-coupling Ising Su-Schrieffer-Heeger (SSH) model. In numerical simulations, the QCNNs achieved remarkable accuracy, correctly identifying known QMBS states with over 99% single-shot measurement accuracy. More importantly, the networks discovered additional non-thermal states beyond the established QMBS families.
For the xorX model specifically, some newly identified non-thermal states can be approximately described as spin-wave modes associated with specific quasiparticles. To better understand these states, the researchers developed effective tight-binding Hamiltonians within the quasiparticle subspace that capture essential features of these many-body eigenstates.
Beyond simulations, the research team validated their approach experimentally using IBM quantum devices. Despite real-world noise and errors, they achieved approximately 63% single-shot measurement accuracy by implementing error mitigation techniques.
This work demonstrates the power of quantum machine learning approaches in uncovering hidden non-thermal states within quantum many-body systems. The success in identifying both known and previously undiscovered non-thermal states underscores the potential of QCNNs as valuable tools for exploring ergodicity breaking and complex quantum phenomena that challenge conventional analytical methods.
Reference: eng, JJ., Zhang, B., Yang, ZC. et al. Uncovering quantum many-body scars with quantum machine learning. npj Quantum Information, Published online: 11 March 2025; doi:10.1038/s41534-025-01005-0
