Multiqubit state learning with entangling quantum generative adversarial networks

The increasing success of classical generative adversarial networks (GANs) has inspired several quantum versions of GANs. Fully quantum mechanical applications of such quantum GANs have been limited to one -and two-qubit systems. In this paper, we investigate the entangling quantum GAN (EQ-GAN) for multiqubit learning. We show that the EQ-GAN can learn a circuit more efficiently compared with a SWAP test. We also consider the EQ-GAN for learning eigenstates that are variational quantum eigensolver (VQE) approximated and find that it generates excellent overlap matrix elements when learning VQE states of small molecules. However, this does not directly translate into a good estimate of the energy due to a lack of phase estimation. Finally, we consider random state learning with the EQ-GAN for up to six qubits, using different two-qubit gates, and show that it is capable of learning completely random quantum states, something which could be useful in quantum state loading.

Automated summary

In this paper, the entangling quantum generative adversarial network (EQ-GAN) is investigated for multiqubit learning. It is shown that EQ-GAN can learn circuits more efficiently than SWAP test and generate excellent overlap matrix elements for learning VQE states of small molecules. However, the lack of phase estimation prevents it from directly estimating energy. Additionally, EQ-GAN demonstrates its potential in learning random states.