Neural quantum states
Bohrdt Group · Research
Neural quantum states
Using restricted Boltzmann machines and other neural network architectures, we perform quantum state reconstruction from measurements. We implement active learning algorithms to optimize measurement selection, maximizing information gain with fewer total measurements needed.
Neural quantum states overcome the exponential scaling of quantum systems by compressing the quantum state in terms of network parameters, rather than storing all exponentially many coefficients. This enables efficient simulation of strongly correlated quantum systems on lattices that are beyond the reach of conventional methods.
A key recent advance is the development of Gutzwiller projected hidden fermion determinant states (G-HFDS), which allow us to simulate the strongly interacting limit of the Fermi-Hubbard model – the t-J model – across the entire doping regime. These states achieve energies competitive with matrix product states on lattices as large as 10×10 sites while using several orders of magnitude fewer parameters.
Selected References
|
Simulating the two-dimensional t-J model at finite doping with neural quantum states
H. Lange, A. Böhler, C. Roth, A. Bohrdt
|
|
Beyond-classical neural quantum states simulation of square-lattice Hubbard model
F. Döschl, F. Palm, H. Lange, F. Grusdt, A. Bohrdt
|
|
Approximately symmetric neural networks for quantum spin liquids
D. S. Kufel, J. Kemp, D. D. Vu, S. M. Linsel, C. R. Laumann, N. Y. Yao
|