Quantum spin liquids (QSLs) are highly elusive states of matter that are char- acterized by long range quantum entanglement and the lack of long range magnetic ordering. There are a variety of candidate materials with QSLs as well as model Hamiltonians where exact QSL ground states can be proven. The Kitaev honeycomb model is among the few models with an analytically derived QSL ground state. Ad- ditional interactions such as the Gamma and Heisenberg interactions are included to match real behaviours in candidate Kitaev materials to form the larger JKΓΓ′model. Due to the size of the Hilbert spaces associated with many-body systems, numerical techniques centre around the use of more efficient representations of the wavefunc- tion that reduce the need for exponential parameterization. Neural Quantum States (NQS) are one of the latest additions to the toolkit, making use of neural networks to represent the mapping between spin configurations and their corresponding wave- function amplitudes. By using variational monte carlo (VMC) to derive the ground state, the problem of finding the associated NQS representation translates to a Ma- chine Learning task of minimizing a cost function through gradient descent. Group Convolutional Neural Networks (GCNNs) are one of the latest NQS architectures which construct wavefunction ansatzes that are symmetric to a group of symmetries and have yielded state-of-the-art results in benchmark models. In this thesis, ground state calculations were performed using GCNNs for the Kitaev and Gamma honey- comb models and ground state properties were investigated. In the Kitaev model, optimization was affected by a complicated learning landscape, limiting the conver- gence towards the ground state. Performance improved with the Gamma model for smaller systems but still fell short of confirming an identical stripy patterned ground state found with iDMRG and iPEPS.
The Kitaev model, which hosts a quantum spin liquid (QSL) in the ground state, was originally defined on a two-dimensional honeycomb lattice, but can be straightforwardly extended to any tricoordinate lattices in any spatial dimensions. In particular, the three-dimensional (3D) extensions are of interest as a realization of 3D QSLs, and some materials like
β-Li
2IrO
3,
γ-Li
2IrO
3, and
β-ZnIrO
3 were proposed for the candidates. However, the phase diagrams of the models for those candidates have not been fully elucidated, mainly due to the limitation of numerical methods for 3D frustrated quantum spin systems. Here we study the Kitaev-Heisenberg model defined on a 3D hyperhoneycomb lattice, by using the pseudofermion functional renormalization group method. We show that the ground-state phase diagram contains the QSL phases in the vicinities of the pristine ferromagnetic and antiferromagnetic Kitaev models, in addition to four magnetically ordered phases, similar to the two-dimensional honeycomb case. Our results respect the four-sublattice symmetry inherent in the model, which was violated in the previous study. Moreover, we also show how the phase diagram changes with the anisotropy in the interactions. The results provide a reference for the search of the hyperhoneycomb Kitaev materials.
We show that neural quantum states based on very deep (4--16-layered) neural networks can outperform state-of-the-art variational approaches on highly frustrated quantum magnets, including quantum-spin-liquid candidates. We focus on group convolutional neural networks (GCNNs) that allow us to impose space-group symmetries on our ansätze. We achieve state-of-the-art ground-state energies for the
J1−J2 Heisenberg models on the square and triangular lattices, in both ordered and spin-liquid phases, and discuss ways to access low-lying excited states in nontrivial symmetry sectors. We also compute spin and dimer correlation functions for the quantum paramagnetic phase on the triangular lattice, which do not indicate either conventional or valence-bond ordering.
Neural network quantum states provide a novel representation of the many-body states of interacting quantum systems and open up a promising route to solve frustrated quantum spin models that evade other numerical approaches. Yet its capacity to describe complex magnetic orders with large unit cells has not been demonstrated, and its performance in a rugged energy landscape has been questioned. Here we apply restricted Boltzmann machines (RBMs) and stochastic gradient descent to seek the ground states of a compass spin model on the honeycomb lattice, which unifies the Kitaev model, Ising model and the quantum 120° model with a single tuning parameter. We report calculation results on the variational energy, order parameters and correlation functions. The phase diagram obtained is in good agreement with the predictions of tensor network ansatz, demonstrating the capacity of RBMs in learning the ground states of frustrated quantum spin Hamiltonians. The limitations of the calculation are discussed. A few strategies are outlined to address some of the challenges in machine learning frustrated quantum magnets.
The Kitaev model, whose ground state is a quantum spin liquid (QSL), was originally conceived for spin
S=1/2 moments on a honeycomb lattice. In recent years, the model has been extended to higher
S from both theoretical and experimental interests, but the stability of the QSL ground state has not been systematically clarified for general
S, especially in the presence of other additional interactions, which inevitably exist in candidate materials. Here we study the spin-
S Kitaev-Heisenberg models by using an extension of the pseudofermion functional renormalization group method to general
S. We show that, similar to the
S=1/2 case, the phase diagram for higher
S contains the QSL phases in the vicinities of the pristine ferromagnetic and antiferromagnetic Kitaev models, in addition to four magnetically ordered phases. We find, however, that the QSL phases shrink rapidly with increasing
S, becoming vanishingly narrow for $S\ensuremath{\ge}2$, whereas the phase boundaries between the ordered phases remain almost intact. Our results provide a reference for the search for higher-
S Kitaev materials.
Abstract A family of spin–orbit coupled honeycomb Mott insulators offers a playground to search for quantum spin liquids (QSLs) via bond-dependent interactions. In candidate materials, a symmetric off-diagonal Γ term, close cousin of Kitaev interaction, has emerged as another source of frustration that is essential for complete understanding of these systems. However, the ground state of honeycomb Γ model remains elusive, with a suggested zigzag magnetic order. Here we attempt to resolve the puzzle by perturbing the Γ region with a staggered Heisenberg interaction which favours the zigzag ordering. Despite such favour, we find a wide disordered region inclusive of the Γ limit in the phase diagram. Further, this phase exhibits a vanishing energy gap, a collapse of excitation spectrum, and a logarithmic entanglement entropy scaling on long cylinders, indicating a gapless QSL. Other quantities such as plaquette-plaquette correlation are also discussed.
In this thesis we study the quantum many-body problem with tools from machine learning. In particular, we utilize Neural Quantum States (NQS) as an ansatz in Variational Monte Carlo (VMC) simulations. First, we develop a method to train NQS more reliably in an effort to reduce the total computational cost of finding the ground state of strongly-correlated systems. Then, we develop an expressive NQS ansatz to study two-dimensional materials and perform ab-initio phase discovery. In particular, we consider the two-dimensional electron gas, and provide state-of-the-art energies and characterization of the phase diagram. We also find evidence of a novel state between the Fermi liquid and Wigner crystal which we call a Nematic Spin Correlated Liquid (NSCL). Lastly, we turn our attention to a moir\'e system with a honeycomb potential. With minor modifications to the ansatz, we find a highly correlated state we call an interaction-driven molecular Wigner crystal. With the increasing interest in NQS as an accurate method to study the many-body problem and recent advances in experimentally studying two-dimensional materials, we expect the technical developments and physical results presented in this thesis to be valuable to theorists and experimentalists.