The Stochastic Gradient Approximation (SGA) is the natural extension of Quantum Monte Carlo (QMC) methods to the variational optimization of quantum wave function parameters. While many deterministic applications impose stochasticity, the SGA fruitfully takes advantage of the natural stochasticity already present in QMC in order to utilize a small number of QMC samples and approach the minimum more quickly by averaging out the random noise in the samples. The increasing efficiency of the method for systems with larger numbers of particles, and its nearly ideal scaling when running on parallelized processors, is evidence that the SGA is well suited for the study of nanoclusters. In this thesis, I discuss the SGA algorithm in detail. I also describe its application to both quantum dots, and to the Resonating Valence Bond wave function (RVB). The RVB is a sophisticated model of electronic systems that captures electronic correlation effects directly and that improves the nodal structure of quantum wave functions. The RVB is receiving renewed attention in the study of nanoclusters due to the fact that calculations of RVB wave functions have become feasible with recent advances in computer hardware and software.
The Stochastic Gradient Approximation (SGA) is the natural extension of Quantum Monte Carlo (QMC) methods to the variational optimization of quantum wave function parameters. While many deterministic applications impose stochasticity, the SGA fruitfully takes advantage of the natural stochasticity already present in QMC in order to utilize a small number of QMC samples and approach the minimum more quickly by averaging out the random noise in the samples. The increasing efficiency of the method for systems with larger numbers of particles, and its nearly ideal scaling when running on parallelized processors, is evidence that the SGA is well suited for the study of nanoclusters. In this thesis, I discuss the SGA algorithm in detail. I also describe its application to both quantum dots, and to the Resonating Valence Bond wave function (RVB). The RVB is a sophisticated model of electronic systems that captures electronic correlation effects directly and that improves the nodal structure of quantum wave functions. The RVB is receiving renewed attention in the study of nanoclusters due to the fact that calculations of RVB wave functions have become feasible with recent advances in computer hardware and software.
Over the past several decades, computational approaches to studying strongly-interacting systems have become increasingly varied and sophisticated. This book provides a comprehensive introduction to state-of-the-art quantum Monte Carlo techniques relevant for applications in correlated systems. Providing a clear overview of variational wave functions, and featuring a detailed presentation of stochastic samplings including Markov chains and Langevin dynamics, which are developed into a discussion of Monte Carlo methods. The variational technique is described, from foundations to a detailed description of its algorithms. Further topics discussed include optimisation techniques, real-time dynamics and projection methods, including Green's function, reptation and auxiliary-field Monte Carlo, from basic definitions to advanced algorithms for efficient codes, and the book concludes with recent developments on the continuum space. Quantum Monte Carlo Approaches for Correlated Systems provides an extensive reference for students and researchers working in condensed matter theory or those interested in advanced numerical methods for electronic simulation.
Quantum Monte Carlo (QMC) is one of the most promising methods for solving quantum many-body problems. QMC combines known analytical properties of wave functions, results from other methods, and stochastic techniques, into a powerful tool for investigation of the electronic structure of real materials. Furthermore, the method has a favorable scaling ( O[N3] in the number of electrons), perfect scalability on parallel architectures, and can be applied to a wide range of systems. Our work has focused both on expanding the capabilities of QMC for calculation of quantities beyond energies, such as interatomic forces, and on application of the method to challenging problems in nanostructure materials research. For the calculation of interatomic forces we have implemented a finite difference correlated sampling method. The correlated sampling enables us to avoid the problem of statistical noise in evaluating the energy differences and allows us to obtain forces with an average 1% error. We have analyzed the effect of the method's main approximations on the estimate of the forces, both by theoretical arguments and by performing a variety of tests on small systems. Furthermore, we have used this newly developed method for estimating equilibrium geometries in a problem where other methods have failed. QMC enabled us to treat the electron correlation effects with high accuracy and to provide accurate predictions in a number of interesting applications. In particular, for carbon rings of C4N+2 stoichiometry we have resolved a long-standing problem regarding the competition between the aromaticity and second order Jahn-Teller effects. We have found that the dominant mechanism at small and intermediate sizes, which stabilizes the bond angle and bond length alternated geometries, is the secondorder Jahn-Teller effect while the aromatic isomer is always found to be a transition state. These high accuracy calculations which have involved multi-reference wave functions, extensive variational optimizations, and the use of massively parallel platforms demonstrate clearly the unique capabilities of the QMC approach.
The quantum Monte Carlo (QMC) is one of the most promising many-body electronic structure approaches. It employs stochastic techniques for solving the stationary Schr\" odinger equation and for evaluation of expectation values. The key advantage of QMC is its capability to use the explicitly correlated wave functions, which allow the study of many-body effects beyond the reach of mean-field methods. The most important limit on QMC accuracy is the fixed-node approximation, which comes from necessity to circumvent the fermion sign problem. The size of resulting fixed-node errors depends on the quality of the nodes (the subset of position space where the wave function vanishes) of a used wave function. In this dissertation, we analyze the nodal properties of the existing fermionic wave functions and offer new types of variational wave functions with improved nodal structure.
Quantum Monte Carlo (QMC) methods are powerful, stochastic techniques for computing the properties of interacting electrons and nuclei with an accuracy comparable to the standard post-Hartree--Fock methods of quantum chemistry. Whilst the favourable scaling of QMC methods enables a quantum, many-body treatment of much larger systems, the lack of accurate and efficient total energy derivatives, required to compute atomic forces, has hindered their widespread adoption. The work contained within this thesis provides an efficient procedure for calculating exact derivatives of QMC results. This procedure uses the programming technique of algorithmic differentiation (AD), which allows access to the derivatives of a computed function by applying chain rule differentiation to the underlying source code. However, this thesis shows that a straightforward differentiation of a stochastic function fails to capture the important contribution to the derivative from probabilistic decisions. A general approach for calculating the derivatives of a stochastic function is presented, where a similar adaptation of AD applied to the diffusion Monte Carlo (DMC) algorithm yields exact DMC atomic forces. The approach is validated by performing the largest ever DMC force calculations, which demonstrate the feasibility of treating systems containing thousands of electrons. The efficiency of AD also enables molecular dynamics simulations driven entirely by DMC, adding new functionality to the QMC toolkit. Another focus of this thesis is using the phenomenon of stochastic coherence to correlate DMC simulations, allowing finite difference derivatives to be obtained with a small error. Whilst this method is far easier to implement than AD, preliminary results show an instability when treating larger systems. A different approach is obtained from extrapolating this method to a finite difference step size of zero, producing algebraic expressions for a direct differentiation of the DMC algorithm.
Electronic structure quantum Monte Carlo Quantum Monte Carlo (QMC) is an advanced simulation methodology for studies of manybody quantum systems. The QMC approaches combine analytical insights with stochastic computational techniques for efficient solution of several classes of important many-body problems such as the stationary Schrödinger equation. QMC methods of various flavors have been applied to a great variety of systems spanning continuous and lattice quantum models, molecular and condensed systems, BEC-BCS ultracold condensates, nuclei, etc. In this review, we focus on the electronic structure QMC, i.e., methods relevant for systems described by the electron-ion Hamiltonians. Some of the key QMC achievements include direct treatment of electron correlation, accuracy in predicting energy differences and favorable scaling in the system size. Calculations of atoms, molecules, clusters and solids have demonstrated QMC applicability to real systems with hundreds of electrons while providing 90-95% of the correlation energy and energy differences typically within a few percent of experiments. Advances in accuracy beyond these limits are hampered by the so-called fixed-node approximation which is used to circumvent the notorious fermion sign problem. Many-body nodes of fermion states and their properties have therefore become one of the important topics for further progress in predictive power and efficiency of QMC calculations. Some of our recent results on the wave function nodes and related nodal domain topologies will be briefly reviewed. This includes analysis of few-electron systems and descriptions of exact and approximate nodes using transformations and projections of the highly-dimensional nodal hypersurfaces into the 3D space. Studies of fermion nodes offer new insights into topological properties of eigenstates such as explicit demonstrations that generic fermionic ground states exhibit the minimal number of two nodal domains. Recently proposed trial wave functions based on Pfaffians with pairing orbitals are presented and their nodal properties are tested in calculations of first row atoms and molecules. Finally, backflow "dressed" coordinates are introduced as another possibility for capturing correlation effects and for decreasing the fixed-node bias.