Determinant- and Derivative-Free Quantum Monte Carlo Within the Stochastic Representation of Wavefunctions
arXiv:2402.06577 · doi:10.1088/1361-6633/ad7d33
Abstract
Describing the ground states of continuous, real-space quantum many-body systems, like atoms and molecules, is a significant computational challenge with applications throughout the physical sciences. Recent progress was made by variational methods based on machine learning (ML) ansatzes. However, since these approaches are based on energy minimization, ansatzes must be twice differentiable. This (a) precludes the use of many powerful classes of ML models; and (b) makes the enforcement of bosonic, fermionic, and other symmetries costly. Furthermore, (c) the optimization procedure is often unstable unless it is done by imaginary time propagation, which is often impractically expensive in modern ML models with many parameters. The stochastic representation of wavefunctions (SRW), introduced in Nat Commun 14, 3601 (2023), is a recent approach to overcoming (c). SRW enables imaginary time propagation at scale, and makes some headway towards the solution of problem (b), but remains limited by problem (a). Here, we argue that combining SRW with path integral techniques leads to a new formulation that overcomes all three problems simultaneously. As a demonstration, we apply the approach to generalized ``Hooke's atoms'': interacting particles in harmonic wells. We benchmark our results against state-of-the-art data where possible, and use it to investigate the crossover between the Fermi liquid and the Wigner molecule within closed-shell systems. Our results shed new light on the competition between interaction-driven symmetry breaking and kinetic-energy-driven delocalization.
References in corpus (46)
- Deep Learning using Rectified Linear Units (ReLU)
- Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems
- Quantum Monte Carlo methods for nuclear physics
- Deep neural network solution of the electronic Schrödinger equation
- Ab-Initio Solution of the Many-Electron Schrödinger Equation with Deep Neural Networks
- Generalized Lanczos Algorithm for Variational Quantum Monte Carlo
- Wigner Crystallization in mesoscopic 2D electron systems
- Wave function optimization in the variational Monte Carlo method
- Crossover from Fermi liquid to Wigner molecule behavior in quantum dots
- Quantum many-body dynamics in two dimensions with artificial neural networks
- Symmetry breaking and quantum correlations in finite systems: Studies of quantum dots and ultracold Bose gases and related nuclear and chemical methods
- Full configuration interaction approach to the few-electron problem in artificial atoms
- NetKet 3: Machine Learning Toolbox for Many-Body Quantum Systems
- Multi-Determinant Wave-functions in Quantum Monte Carlo
- Diffusion Monte Carlo study of circular quantum dots
- Empowering deep neural quantum states through efficient optimization
- Learning the ground state of a non-stoquastic quantum Hamiltonian in a rugged neural network landscape
- Ab initio calculation of real solids via neural network ansatz
- Towards the ground state of molecules via diffusion Monte Carlo on neural networks
- Incipient Wigner Localization in Circular Quantum Dots
- A simple linear algebra identity to optimize Large-Scale Neural Network Quantum States
- Wave function Ansatz (but Periodic) Networks and the Homogeneous Electron Gas
- Iterative backflow renormalization procedure for many-body ground state wave functions of strongly interacting normal Fermi liquids
- Gauge Invariant and Anyonic Symmetric Transformer and RNN Quantum States for Quantum Lattice Models
- Unbiasing time-dependent Variational Monte Carlo by projected quantum evolution
- Nonlinear Network description for many-body quantum systems in continuous space
- Message-Passing Neural Quantum States for the Homogeneous Electron Gas
- Strongly correlated wave functions for artificial atoms and molecules
- Neural-network states for the classical simulation of quantum computing
- High-order Path Integral Monte Carlo methods for solving quantum dot problems
- Rapidly rotating boson molecules with long or short range repulsion: an exact diagonalization study
- Interaction-Induced Strong Localization in Quantum Dots
- A Self-Attention Ansatz for Ab-initio Quantum Chemistry
- Impurity effects in few-electron quantum dots: Incipient Wigner molecule regime
- Ab-initio study of interacting fermions at finite temperature with neural canonical transformation
- DeepQMC: an open-source software suite for variational optimization of deep-learning molecular wave functions
- Harmonically trapped two-atom systems: Interplay of short-range s-wave interaction and spin-orbit coupling
- Convergence to the fixed-node limit in deep variational Monte Carlo
- Quantum Wigner molecules in moiré materials
- Learning ground states of gapped quantum Hamiltonians with Kernel Methods
- Wigner-molecule supercrystal in transition-metal dichalcogenide moiré superlattices: Lessons from the bottom-up approach
- Scalable Imaginary Time Evolution with Neural Network Quantum States
- Energy spectra of two interacting fermions with spin-orbit coupling in a harmonic trap
- Universal Antisymmetry in Fermionic Neural Networks
- Solving fermion problems without solving the sign problem: symmetry-breaking wave functions from similarity-transformed propagators for solving 2D quantum dots
- Sorting Out Quantum Monte Carlo