Solving the fractional quantum Hall problem with self-attention neural network

Yi Teng, David D. Dai, Liang Fu · 2024

Read Solving the fractional quantum Hall problem with self-attention neural network by Yi Teng, David D. Dai, Liang Fu (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-00618.txt.

Introduction

Yi Teng, David Dai, and Liang Fu solve and visualize fractional quantum Hall wavefunctions with a self-attention neural network. The quantum Hall problem, electrons on a disk in a strong magnetic field, hosts exotic liquid and crystal states, but exact solutions choke on the huge state space. Their fermionic network works with the full space directly.

It reaches lower energies than the standard projected exact method and learns the ground state accurately. At weak mixing between Landau levels it reveals fine structure beyond the classic Laughlin picture. At stronger mixing it pulls further ahead and uncovers a transition from liquid to crystal. Attention networks may become standard tools for strongly correlated matter.

Catalog record

CREATORYi Teng, David D. Dai, Liang Fu
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00618

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Solving the fractional quantum Hall problem with self-attention neural network Yi Teng, David D. Dai, Liang Fu Abstract We introduce an attention-based fermionic neural network (FNN) to variationally solve the problem of two-dimensional Coulomb electron gas in magnetic fields, a canonical platform for fractional quantum Hall (FQH) liquids, Wigner crystals and other unconventional electron states. Working directly with the full Hilbert space of N N electrons confined to a disk, our FNN consistently attains energies lower than LL-projected exact diagonalization (ED) and learns the ground state wavefunction to high accuracy. In low LL mixing regime, our FNN reveals microscopic features in the short-distance behavior of FQH wavefunction beyond the Laughlin ansatz. For moderate and strong LL mixing parameters, the FNN outperforms ED significantly. Moreover, a phase transition from FQH liquid to a crystal state is found at strong LL mixing. Our study demonstrates unprecedented power and universality of FNN based variational method for solving strong-coupling many-body problems with topological order and electron fractionalization. I I. Introduction Theoretical quantum condensed matter physics has been driven by the desire to understand and predict complex collective phenomena—such as superconductivity and magnetism—that emerge from interactions between a large number of constituent particles. Due to the exponential growth of Hilbert space dimension with the system size, it is in general not possible to obtain numerically exact solution in the thermodynamic limit. A variety of approximate numerical methods have been developed to study different types of quantum many-body systems. The density-functional theory is remarkably successful for studying real materials with weak electron correlation effects [ 1 , 2 , 3 ] . For strongly correlated electron systems, variational methods based on matrix product states have proven accurate for (quasi) one-dimensional lattice models [ 4 , 5 , 6 ] . In recent years, a wide array of correlated and topological electron states have been discovered in moiré materials, including spin-polaron metals [ 7 , 8 , 9 ] and fractional Chern insulators [ 10 , 11 , 12 , 13 , 14 , 15 ] . As these ground states cannot be adiabatically connected to any mean-field states, theoretical progress has largely come from exact diagonalization (ED) studies, which are limited by small system sizes and truncation to the lowest one or few bands [ 16 , 17 , 18 ] . To advance precision many-body theory, it is highly desirable to develop more powerful and versatile methods for solving strongly interacting electron problems in continuum systems. Recent advances in artificial intelligence and deep learning have motivated the development of neural network-variational Monte Carlo (NN-VMC) method [ 19 , 20 ] . While NN-VMC approach was first introduced and developed for lattice models [ 21 , 22 , 23 , 24 , 25 , 26 ] , an important breakthrough came in continuum systems when fermionic neural networks (FNN) based on generalized Slater determinants, in particular PauliNet [ 27 ] and FermiNet [ 28 , 29 ] , were developed to solve quantum chemistry problems in atoms and molecules with high accuracy. The NN-VMC approach has since been adapted to extended electron systems in continuous space, including real solids [ 30 ] , uniform electron gas [ 31 , 32 , 33 ] and moiré materials [ 34 , 35 ] . As a sign-problem-free variational method, NN-VMC provides a variational upper bound on the ground state energy, and its computational cost scales polynomially with the system size [ 36 , 37 ] . Thanks to the remarkable expressive power of neural network wavefunctions, this method can in principle solve different phases using a unified and unbiased ansatz. So far, NN-VMC has proven successful in studying Fermi liquid [ 31 , 32 ] and Wigner crystals [ 31 , 33 ] . On the other hand, it remains an open question whether the standard NN architecture (such as FermiNet) is capable of simulating exotic quantum states beyond mean-field description, such as fractionalized quantum phases of matter. In this work, we adapt the self-attention FNN Psiformer [ 36 ] to study two-dimensional electron gas in strong magnetic fields—a canonical setting and enduring source for strongly correlated electron phases such as fractional quantum Hall states and composite Fermi liquids. Using NN-VMC, we are able to solve this many-electron problem directly in real space , without using Landau level (LL) projection and truncation. Our FNN finds the \nu=1/3 fractional quantum Hall state (FQH) over a range of LL mixing parameters and consistently attains energies lower than LL-projected ED. The final FNN wavefunction exhibits phase structure and pair distribution at short distance that are clearly distinct from the Laughlin ansatz. For realistic parameters, our FNN finds substantial LL mixing in the wavefunction and significantly outperforms LL-projected ED. At very strong LL mixing parameter, we discover a phase transition from the FQH liquid to another state with distinct charge density profile and quantum number. Our work demonstrates the power and versatility of FNN architecture for studying fractionalized quantum matter in continuous space and solving real-world solid-state problems. Our study unlocks the immense potential of NN-VMC in tackling strong correlation and topological order with a unifying and unbiased ansatz, opening the door to new applications of AI in quantum condensed matter physics. II II. Laughlin wavefunction and FQH ground state The standard approach to the FQH problem assumes that electron-electron interaction is weak compared to the cyclotron gap, so that LL mixing is neglected to the zeroth order approximation. In his groundbreaking work [ 38 ] , Laughlin proposed an elegant many-body wavefunction within the lowest Landau level (LLL) subspace at fractional filling \nu=1/m ( m odd): \Psi_{L}=\prod_{\{i,j\}}(z_{i}-z_{j})^{m}e^{-{\Sigma_{i}\lvert z_{i}\lvert^{2}}/{4\ell_{B}^{2}}}\,, (1) where z=x+iy is the standard complex coordinate, and \ell_{B} is the magentic length. While the Laughlin wavefunction captures the essential physics of FQH liquids [ 38 , 39 ] , it is not the exact ground state of 2D Coulomb electron gas under magnetic field. The key difference lies in two aspects. First, in the Laughlin wavefunction the relative angular momentum l of two electrons only admits l\geq m components, and the zeros are all m^{\rm th} order. Under realistic interaction including Coulomb, there will be non-zero components for l<m , with l=1 component the most dominant at short distance <\ell_{B} . As a result, each m^{\rm th} order zero of the wavefunction splits into m first-order zeros [ 40 , 41 , 42 , 43 , 44 ] . Moreover, for many semiconductor materials under realistic magnetic fields [ 45 , 46 ] , the interaction energy scale is comparable to or (considerably) larger than the cyclotron gap. In this case, LL mixing becomes significant, and taking into account higher LL levels is of quantitative or even qualitative importance. Historically, theoretical advances on the FQH problem have benefited greatly from various numerical methods including ED [ 47 , 48 , 49 , 50 , 51 , 52 ] , density matrix renormalization group [ 53 ] and fixed-phase diffusion Monte Carlo (DMC) [ 54 , 55 , 56 , 57 ] . Despite their great success, these methods have inherent limitations. ED and DMRG are limited to relatively small system size and quasi-one-dimensional geometry respectively. Moreover, both methods are limited by

Read the complete text