DOCSLIB.ORG

Arxiv:1902.04057V3 [Cond-Mat.Dis-Nn] 19 Jan 2020 Statistical Mechanics Applications [16], As Well As Den- Spin Systems [13]

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Arxiv:1902.04057V3 [Cond-Mat.Dis-Nn] 19 Jan 2020 Statistical Mechanics Applications [16], As Well As Den- Spin Systems [13] Deep autoregressive models for the efficient variational simulation of many-body quantum systems 1, 1, 1, 2, 1, Or Sharir, ∗ Yoav Levine, † Noam Wies, ‡ Giuseppe Carleo, § and Amnon Shashua ¶ 1The Hebrew University of Jerusalem, Jerusalem, 9190401, Israel 2Center for Computational Quantum Physics, Flatiron Institute, 162 5th Avenue, New York, NY 10010, USA Artificial Neural Networks were recently shown to be an efficient representation of highly-entangled many-body quantum states. In practical applications, neural-network states inherit numerical schemes used in Variational Monte Carlo, most notably the use of Markov-Chain Monte-Carlo (MCMC) sampling to estimate quantum expectations. The local stochastic sampling in MCMC caps the potential advantages of neural networks in two ways: (i) Its intrinsic computational cost sets stringent practical limits on the width and depth of the networks, and therefore limits their expressive capacity; (ii) Its difficulty in generating precise and uncorrelated samples can result in estimations of observables that are very far from their true value. Inspired by the state-of-the-art generative models used in machine learning, we propose a specialized Neural Network architecture that supports efficient and exact sampling, completely circumventing the need for Markov Chain sampling. We demonstrate our approach for two-dimensional interacting spin models, showcas- ing the ability to obtain accurate results on larger system sizes than those currently accessible to neural-network quantum states. Introduction.– The theoretical understanding and been limited to relatively shallow architectures, far from modeling of interacting many-body quantum matter rep- the very deep networks used in modern machine learn- resents an outstanding challenge since the early days of ing applications. This practical limitation is mostly due quantum mechanics. At the heart of several problems in to two main factors. First, it is computationally expen- condensed matter, chemistry, nuclear matter, and more sive to obtain quantum expectation values over ConvNet lies the intrinsic difficulty of fully representing the many- states using stochastic sampling based on Markov Chain body wave-function, in principle needed to exactly solve Monte Carlo (MCMC), as is it is customary in VMC Schrodinger’s equation. These mainly fall into two cat- applications. Second, there is an intrinsic optimization egories: on one hand, there are states traditionally used bottleneck to be faced when dealing with a large number in stochastic Variational Monte Carlo (VMC) calcula- of parameters. However, both limitations are routinely tions [1]. Chief example are Jastrow wave-functions [2], faced when learning deep autoregressive-models, recently carrying high entanglement, but also with a limited varia- introduced machine-learning techniques that have en- tional freedom. On the other hand, more recently, tensor- abled previously intractable applications. network approaches have been put forward, based on In this paper, we propose a pivotal shift in the use of non-stochastic variational optimization, and most chiefly Neural-network Quantum States (NQS) for many-body on entanglement-limited variational wave-functions [3–6]. quantum systems, that markedly sets a discontinuity In an attempt to circumvent the limitations of the ap- with traditionally adopted VMC methods. Inspired by proaches above, architectures based on Artificial Neural the latest advances in generative machine learning mod- Networks (ANN) were proposed as variational wave func- els, we introduce variational states for which both the tions [7]. Restricted Boltzmann machines (RBM), which sampling and the optimization issues are substantially represent relatively veteran machine learning constructs, alleviated. Our model is composed of a ConvNet that were shown to be capable of representing volume-law en- allows direct, efficient, and i.i.d. sampling from the tanglement scaling in 2D [8–11]. Recently, other neural- highly entangled wave function it represents. The net- network architectures have been explored. Most notably, work architecture draws upon successful autoregressive convolutional neural networks (ConvNets) – leading deep models for representing and sampling from probability learning architectures that stand at the forefront of em- distributions. Those are widely employed in the machine pirical successes in various Artificial Intelligence domains learning literature [15], and have been recently used for – have been applied to both bosonic [12] and frustrated arXiv:1902.04057v3 [cond-mat.dis-nn] 19 Jan 2020 statistical mechanics applications [16], as well as den- spin systems [13]. sity matrix reconstructions from experimental quantum Despite the provable theoretical advantage of ConvNet systems [17]. We generalize these autoregressive models architectures [14], however, early numerical studies have to treat complex-valued wave-functions, obtaining highly expressive architectures parametrizing an automatically normalized many-body quantum wave-function. Neural Autoregressive Quantum States.– We consider ∗ [email protected] † [email protected] in the following a pure quantum system, constituted by [email protected] N discrete degrees of freedom s (s , . , s )(e.g. spins, ‡ ≡ 1 N § gcarleo@flatironinstitute.org occupation numbers, etc.) such that the wave-function ¶ [email protected] amplitudes Ψ(s) fully specify its state. Here we follow 2 the approach introduced in [7], and represent ln(Ψ(s)) probabilities for si to take one of the M possible dis- as a feed-forward ANN, parametrized by a possibly large crete values sj, conditioned on given s1, . , si 1. It is − number of network connections. Given an arbitrary set crucial that each output vector vi does not depend on of quantum numbers, s, the output value computation of the value of si or any of the variables appearing with the corresponding NQS, known as its forward pass, can a larger index, si+1, . , sN , for a pre-chosen ordering. generally be described as a sequence of K matrix-vector To ensure that each network outputs a valid conditional multiplications separated by the applications of a non- distribution, it is then sufficient to take the exponent linear element-wise activation function σ:C C. More of each entry and normalizing it according to the l1 formally, the unnormalized log amplitudes are→ given by norm, i.e. p (s s , . , s )=exp(v )/ exp(v ) , i i i 1 1 i,si s0 i,s0 also known as a|Softmax− operation. | | P ln(Ψ(s)) = WK σ (Wk 1σ ( σ (W1s))) , (1) Even though it is possible to use N separate networks − ··· for each of the N conditional probabilities, and each ac- K ri ri 1 cepting a variable number of inputs, in practice it is where Wi C × − i=1, r0=N, rK =1, r1, . , rK 1 are knownW≡ { as the∈ widths}of the network, and K as the− more common to use a single ANN that accepts N in- depth. In practice, specialized variants of eq.1 are com- puts and outputs N probability vectors. In this case, monly used, e.g. early applications have focused on shal- the autoregressive property is enforced by masking the low architectures (k=1) such as Restricted Boltzmann inputs si, . , sN for the i’th output vector, i.e. ensur- Machines, for which the activation function is typically ing that the contributions of higher-ordered spins to the taken to be σ(z)= ln cosh(z). Other, deeper, choices are output of the network vanish. PixelCNN [18] is such an often advantageous, such as convolutional networks, in architecture, and is built as a sequence of masked convo- which most of the matrices are restricted to act on a lutional layers, whose filters are restricted to having zeros subset of the quantum numbers, computing convolutions at positions “ahead”. For example, in a one dimensional with small filters. system, a filter of width R, where R is odd, would be con- strained to have (w , . , w(R 1) , 0,..., 0), and thus the Given a NQS representation of a many-body quan- 1 − /2 tum state, estimating physical observables Ψ Ψ ith output of each layer depends uniquely on the indices h |O| i at s1, . , si 1. of a local operator , is in general analytically in- − tractable, but can beO realized numerically through a A chief advantage of networks with the autoregressive stochastic procedure, as done in VMC. Specifically, property, is that directly drawing samples according to Ψ Ψ = Oloc , where ... denote statistical ex- P (s) is conceptually straightforward. One can sample h |O| i h iP h iP pectation values over the Born probability density each si in sequence, according to its given conditional 2 loc (s) Ψ(s) , and O s s0 Ψ(s0)/Ψ(s) is the probability that depends just on the previously sampled s0 P ≡ | | ≡ h |O| i (s1, . , si 1). Carefully exploiting the intrinsic sparse- corresponding statistical estimator. In the vast majority − of VMC applications, includingP NQS so-far, a MCMC al- ness of the network weights, further leads to a very effi- gorithm is typically used to generate samples from (s). cient algorithm for sampling [19]. Remarkably, the com- P While MCMC is a rather flexible technique, it comes with plexity of sampling a full string s1 . sN in a PixelCNN a large computational cost, especially for deep ANNs. architecture can be reduced to the complexity of just a Additionally, though MCMC asymptotically generates single forward pass. samples that are correctly distributed, in practice it can Our NAQS model for representing wave-functions is be plagued by very large autocorrelation times, and lack based on the same NADE principles so-far described. of ergodicity, that can severely affect the quality of the Specifically, just as probability functions can be factor- samples being generated. ized into a product of conditional probabilities, we repre- In light of these limitations, we propose here a sent a normalized wave-function as a product of normal- specialized network architecture that instead supports ized conditional wave-functions, such that efficient and exact sampling. Our approach is an extension of Neural Autoregressive Density Estima- N Ψ(s1, .
Load more

Recommended publications
  • An Overview of Quantum Monte Carlo Methods David M. Ceperley
    An Overview of Quantum Monte Carlo Methods David M. Ceperley Department of Physics and National Center for Supercomputing Applications University of Illinois Urbana-Champaign Urbana, Illinois 61801 In this brief article, I introduce the various types of quantum Monte Carlo (QMC) methods, in particular, those that are applicable to systems in extreme regimes of temperature and pressure. We give references to longer articles where detailed discussion of applications and algorithms appear. MOTIVATION One does QMC for the same reason as one does classical simulations; there is no other method able to treat exactly the quantum many-body problem aside from the direct simulation method where electrons and ions are directly represented as particles, instead of as a “fluid” as is done in mean-field based methods. However, quantum systems are more difficult than classical systems because one does not know the distribution to be sampled, it must be solved for. In fact, it is not known today which quantum problems can be “solved” with simulation on a classical computer in a reasonable amount of computer time. One knows that certain systems, such as most quantum many-body systems in 1D and most bosonic systems are amenable to solution with Monte Carlo methods, however, the “sign problem” prevents making the same statement for systems with electrons in 3D. Some limitations or approximations are needed in practice. On the other hand, in contrast to simulation of classical systems, one does know the Hamiltonian exactly: namely charged particles interacting with a Coulomb potential. Given sufficient computer resources, the results can be of quite high quality and for systems where there is little reliable experimental data.
    [Show full text]
  • Variational Monte Carlo Technique Ground State Energies of Quantum Mechanical Systems
    GENERAL ⎜ ARTICLE Variational Monte Carlo Technique Ground State Energies of Quantum Mechanical Systems Sukanta Deb In this article, I discuss the use of the Variational Monte Carlo technique in the determination of ground state energies of quantum harmonic os- cillators in one and three dimensions. In order to provide a better understanding of this technique, a pre-requisite concept of Monte Carlo integra- Sukanta Deb is an tion is also discussed along with numerical ex- Assistant Professor in the amples. The technique can be easily extended Department of Physics at Acharya Narendra Dev to study the ground state energies of hydrogen College (University of atom, particle in a box, helium atom and other Delhi). His research complex microscopic systems involving N parti- interests include astro- cles in d-dimensions. nomical photometry and spectroscopy, light curve 1. Introduction modeling in astronomy and astrophysics, Monte Carlo There exist many problems in science and engineering methods and simulations whose exact solution either does not exist or is difficult and automated data to find. For the solutions of those problems, one has to analysis. resort to approximate methods. In the context of quan- tum mechanics, the exact solutions of the Schr¨odinger’s equation exist only for a few idealized systems. There are varieties of systems for which the Schr¨odinger’s equa- tion either cannot be solved exactly or it involves very lengthy and cumbersome calculations [1]. For example, solving the Schr¨odinger equation to study the proper- ties of a system with hundreds or thousands of particles often turns out to be impractical [2].
    [Show full text]
  • Unifying Cps Nqs
    Citation for published version: Clark, SR 2018, 'Unifying Neural-network Quantum States and Correlator Product States via Tensor Networks', Journal of Physics A: Mathematical and General, vol. 51, no. 13, 135301. https://doi.org/10.1088/1751- 8121/aaaaf2 DOI: 10.1088/1751-8121/aaaaf2 Publication date: 2018 Document Version Peer reviewed version Link to publication This is an author-created, in-copyedited version of an article published in Journal of Physics A: Mathematical and Theoretical. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The Version of Record is available online at http://iopscience.iop.org/article/10.1088/1751-8121/aaaaf2/meta University of Bath Alternative formats If you require this document in an alternative format, please contact: [email protected] General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Download date: 04. Oct. 2021 Unifying Neural-network Quantum States and Correlator Product States via Tensor Networks Stephen R Clarkyz yDepartment of Physics, University of Bath, Claverton Down, Bath BA2 7AY, U.K. zMax Planck Institute for the Structure and Dynamics of Matter, University of Hamburg CFEL, Hamburg, Germany E-mail: [email protected] Abstract.
    [Show full text]
  • An Introduction to Quantum Monte Carlo
    An introduction to Quantum Monte Carlo Jose Guilherme Vilhena Laboratoire de Physique de la Matière Condensée et Nanostructures, Université Claude Bernard Lyon 1 Outline 1) Quantum many body problem; 2) Metropolis Algorithm; 3) Statistical foundations of Monte Carlo methods; 4) Quantum Monte Carlo methods ; 4.1) Overview; 4.2) Variational Quantum Monte Carlo; 4.3) Diffusion Quantum Monte Carlo; 4.4) Typical QMC calculation; Outline 1) Quantum many body problem; 2) Metropolis Algorithm; 3) Statistical foundations of Monte Carlo methods; 4) Quantum Monte Carlo methods ; 4.1) Overview; 4.2) Variational Quantum Monte Carlo; 4.3) Diffusion Quantum Monte Carlo; 4.4) Typical QMC calculation; 1 - Quantum many body problem Consider of N electrons. Since m <<M , to a good e N approximation, the electronic dynamics is governed by : Z Z 1 2 1 H =− ∑ ∇ i −∑∑ ∑∑ 2 i i ∣r i−d∣ 2 i j ≠i ∣ri−r j∣ which is the Born Oppenheimer Hamiltonian. We want to find the eigenvalues and eigenvectos of this hamiltonian, i.e. : H r1, r 2,. ..,r N =E r 1, r2,. ..,r N QMC allow us to solve numerically this, thus providing E0 and Ψ . 0 1 - Quantum many body problem Consider of N electrons. Since m <<M , to a good e N approximation, the elotonic dynamics is governed by : Z Z 1 2 1 H =− ∑ ∇ i −∑∑ ∑∑ 2 i i ∣r i−d∣ 2 i j ≠i ∣ri−r j∣ which is the Born Oppenheimer Hamiltonian. We want to find the eigenvalues and eigenvectos of this hamiltonian, i.e. : H r1, r 2,.
    [Show full text]
  • University of Bath
    Citation for published version: Clark, SR 2018, 'Unifying Neural-network Quantum States and Correlator Product States via Tensor Networks', Journal of Physics A: Mathematical and General, vol. 51, no. 13, 135301. https://doi.org/10.1088/1751- 8121/aaaaf2 DOI: 10.1088/1751-8121/aaaaf2 Publication date: 2018 Document Version Peer reviewed version Link to publication This is an author-created, in-copyedited version of an article published in Journal of Physics A: Mathematical and Theoretical. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The Version of Record is available online at http://iopscience.iop.org/article/10.1088/1751-8121/aaaaf2/meta University of Bath General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Download date: 23. May. 2019 Unifying Neural-network Quantum States and Correlator Product States via Tensor Networks Stephen R Clarkyz yDepartment of Physics, University of Bath, Claverton Down, Bath BA2 7AY, U.K. zMax Planck Institute for the Structure and Dynamics of Matter, University of Hamburg CFEL, Hamburg, Germany E-mail: [email protected] Abstract. Correlator product states (CPS) are a powerful and very broad class of states for quantum lattice systems whose (unnormalised) amplitudes in a fixed basis can be sampled exactly and efficiently.
    [Show full text]
  • Variational Monte Carlo Method for Fermions
    Variational Monte Carlo Method for Fermions Vijay B. Shenoy [email protected] Indian Institute of Science, Bangalore December 26, 2009 1/91 Acknowledgements Sandeep Pathak Department of Science and Technology 2/91 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas Details of VMC Application to Superconductivity 3/91 The Strongly Correlated Panorama... Fig. 1. Phase diagrams ofrepresentativemateri- als of the strongly cor- related electron family (notations are standard and details can be found in the original refer- ences). (A) Temperature versus hole density phase diagram of bilayer man- ganites (74), including several types of antiferro- magnetic (AF) phases, a ferromagnetic (FM) phase, and even a glob- ally disordered region at x 0 0.75. (B) Generic phase diagram for HTSC. SG stands for spin glass. (C) Phase diagram of single layered ruthenates (75, 76), evolving from a superconducting (SC) state at x 0 2 to an AF insulator at x 0 0(x controls the bandwidth rather than the carrier density). Ruthenates are believed to be clean metals at least at large x, thus providing a family of oxides where competition and com- plexity can be studied with less quenched dis- order than in Mn ox- ides. (D) Phase diagram of Co oxides (77), with SC, charge-ordered (CO), and magnetic regimes. (E) Phase diagram of the organic k-(BEDT- TTF)2Cu[N(CN)2]Cl salt (57). The hatched re- gion denotes the co- existence of metal and insulator phases. (F) Schematic phase dia- gram of the Ce-based heavy fermion materials (51).
    [Show full text]
  • Quantum Monte Carlo Simulations: from Few to Many Species
    Quantum Monte Carlo Simulations: From Few to Many Species by William Gary Dawkins A Thesis Presented to The University of Guelph In partial fulfilment of requirements for the degree of Master of Science in Physics Guelph, Ontario, Canada c William Gary Dawkins, May, 2019 ABSTRACT QUANTUM MONTE CARLO SIMULATIONS: FROM FEW TO MANY SPECIES William Gary Dawkins Advisor: University of Guelph, 2019 Professor Alexandros Gezerlis We study systems of particles with varying numbers of distinguishable species ranging from 4-species to the N-species limit, where N is the number of particles in the system itself. We examine these systems using various powerful simulation techniques that belong to the quantum Monte Carlo (QMC) family. The N-species limit is studied via Path Integral Monte Carlo (PIMC), which is a finite temperature, non- perturbative technique. At this limit, we explore the viability of two different thermal density matrices for systems that interact via hard-sphere, hard-cavity potentials. We then study the impact of finite-size effects on calculations of the energy, pressure and specific heat of the system when reaching the thermodynamic limit under periodic boundary conditions. We then move to studying a 4-species fermionic system using Variational Monte Carlo (VMC) and Diffusion Monte Carlo (DMC), which are ground state techniques. We apply these methods to the clustering problem of 4-particle and 8-particle systems. While the 4-particle, 4-species system is relatively simple to calculate as it is physically identical to a bosonic system of 4-particles, the 8-particle fermi system is more complicated. We detail the process of simulating a bound 8- particle state with respect to decay into two independent 4-particle clusters with DMC for the first time.
    [Show full text]
  • Deep Learning-Enhanced Variational Monte Carlo Method for Quantum Many-Body Physics
    PHYSICAL REVIEW RESEARCH 2, 012039(R) (2020) Rapid Communications Deep learning-enhanced variational Monte Carlo method for quantum many-body physics Li Yang ,1,* Zhaoqi Leng,2 Guangyuan Yu,3 Ankit Patel,4,5 Wen-Jun Hu,1,† and Han Pu1,‡ 1Department of Physics and Astronomy & Rice Center for Quantum Materials, Rice University, Houston, Texas 77005, USA 2Department of Physics, Princeton University, Princeton, New Jersey 08544, USA 3Department of Physics and Astronomy & The Center for Theoretical Biological Physics, Rice University, Houston, Texas 77005, USA 4Department of Electrical and Computer Engineering, Rice University, Houston, Texas 77005, USA 5Department of Neuroscience, Baylor College of Medicine, Houston, Texas 77030, USA (Received 26 May 2019; accepted 12 December 2019; published 14 February 2020) Artificial neural networks have been successfully incorporated into the variational Monte Carlo method (VMC) to study quantum many-body systems. However, there have been few systematic studies exploring quantum many-body physics using deep neural networks (DNNs), despite the tremendous success enjoyed by DNNs in many other areas in recent years. One main challenge of implementing DNNs in VMC is the inefficiency of optimizing such networks with a large number of parameters. We introduce an importance sampling gradient optimization (ISGO) algorithm, which significantly improves the computational speed of training DNNs by VMC. We design an efficient convolutional DNN architecture to compute the ground state of a one-dimensional SU(N) spin chain. Our numerical results of the ground-state energies with up to 16 layers of DNNs show excellent agreement with the Bethe ansatz exact solution. Furthermore, we also calculate the loop correlation function using the wave function obtained.
    [Show full text]
  • Variational Quantum Monte Carlo Abstract Introduction
    Variational Quantum Monte Carlo Josh Jiang, The University of Chicago > (1.1) Abstract The many electron system is one that quantum mechanics have no exact solution for because of the correlation between each electron with every other electron. Many methods have been developed to account for this correlation. One family of methods known as quantum Monte Carlo methods takes a numerical approach to solving the many-body problem. By employing a stochastic process to estimate the energy of a system, this method can be used to calculate the ground state energy of a system using the variational principle. The core of this method involves choosing an ansatz with a set of variational parameters that can be optimized to minimize the energy of the system. Recent research in this field is focused on developing methods to faster and more efficiently optimize this trial ansatz and also to make the ansatz more flexible by allowing more parameters. These topics along with the theory behind variational Monte Carlo (VMC) will be the subject of this Maple worksheet. Introduction Quantum mechanics is the study of particles at a very small scale. At this scale, particles can be characterized by wave functions and behave much differently than classical systems. By understanding how molecules behave at the quantum level, researchers can apply that knowledge to gain insights into how macroscopic systems work. While the applications are promising, quantum mechanics still struggles to understand the behavior of large atomic systems. For small, simple systems like particle in a box, harmonic oscillator, and the hydrogen atom, the Schrodinger equation: can be solved exactly, revealing the energy levels and exact wave functions for those systems.
    [Show full text]
  • Variational Monte Carlo Method for Fermions
    Variational Monte Carlo Method for Fermions Vijay B. Shenoy [email protected] Indian Institute of Science, Bangalore Mahabaleswar, December 2009 1/36 Acknowledgements Sandeep Pathak 2/36 Acknowledgements Sandeep Pathak Department of Science and Technology 2/36 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas (Lecture 1, VBS) 3/36 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas (Lecture 1, VBS) Details of VMC (Lecture 2, VBS) 3/36 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas (Lecture 1, VBS) Details of VMC (Lecture 2, VBS) Application to Superconductivity (Lecture 3, VBS) 3/36 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas (Lecture 1, VBS) Details of VMC (Lecture 2, VBS) Application to Superconductivity (Lecture 3, VBS) Application to Bosons (Lecture 4, AP) 3/36 Plan of the Lectures on Variational Monte Carlo Method Motivation, Overview and Key Ideas (Lecture 1, VBS) Details of VMC (Lecture 2, VBS) Application to Superconductivity (Lecture 3, VBS) Application to Bosons (Lecture 4, AP) Application to Spin Liquids (Lecture 5, AP) 3/36 The Strongly Correlated Panorama... Fig. 1. Phase diagrams ofrepresentativemateri- als of the strongly cor- related electron family (notations are standard and details can be found in the original refer- ences). (A) Temperature versus hole density phase diagram of bilayer man- ganites (74), including several types of antiferro- magnetic (AF) phases, a ferromagnetic (FM) phase, and even a glob- ally disordered region at x 0 0.75.
    [Show full text]
  • Arxiv:1807.10633V1 [Physics.Chem-Ph] 27 Jul 2018
    Improved speed and scaling in orbital space variational monte carlo Iliya Sabzevari1 and Sandeep Sharma1, ∗ 1Department of Chemistry and Biochemistry, University of Colorado Boulder, Boulder, CO 80302, USA In this work, we introduce three algorithmic improvements to reduce the cost and improve the scal- ing of orbital space variational Monte Carlo (VMC). First, we show that by appropriately screening the one- and two-electron integrals of the Hamiltonian one can improve the efficiency of the al- gorithm by several orders of magnitude. This improved efficiency comes with the added benefit that the resulting algorithm scales as the second power of the system size O(N 2), down from the fourth power O(N 4). Using numerical results, we demonstrate that the practical scaling obtained is in fact O(N 1:5) for a chain of Hydrogen atoms, and O(N 1:2) for the Hubbard model. Second, we introduce the use of the rejection-free continuous time Monte Carlo (CTMC) to sample the determinants. CTMC is usually prohibitively expensive because of the need to calculate a large number of intermediates. Here, we take advantage of the fact that these intermediates are already calculated during the evaluation of the local energy and consequently, just by storing them one can use the CTCM algorithm with virtually no overhead. Third, we show that by using the adaptive stochastic gradient descent algorithm called AMSGrad one can optimize the wavefunction energies robustly and efficiently. The combination of these three improvements allows us to calculate the ground state energy of a chain of 160 hydrogen atoms using a wavefunction containing ∼ 2 × 105 variational parameters with an accuracy of 1 mEh/particle at a cost of just 25 CPU hours, which when split over 2 nodes of 24 processors each amounts to only about half hour of wall time.
    [Show full text]
  • Variational Monte Carlo Algorithm for Solving One Dimensional Harmonic Oscillator Problem the Journal of Mathematics and Comput
    Farshid Mehrdoust, Hossein Aminikhah, Mohammad Ghamgosar / TJMCS Vol .3 No.4 (2011) 376 - 381 The Journal of Mathematics and Computer Science Available online at http://www.TJMCS.com The Journal of Mathematics and Computer Science Vol .3 No.4 (2011) 376 - 381 Variational Monte Carlo algorithm for solving one dimensional harmonic oscillator problem Farshid Mehrdoust Department of Applied Mathematics, Faculty of Mathematical Sciences, University of Guilan, Rasht, Iran, [email protected] Hossein Aminikhah Department of Applied Mathematics, Faculty of Mathematical Sciences, University of Guilan, Rasht, Iran, [email protected] Mohammad Ghamgosar [email protected] Received: September 2011, Revised: November 2011 Online Publication: December 2011 Abstract The goal of this paper is to present an application of variational Monte Carlo method for solving one dimensional harmonic oscillator problem. Keywords: Variational Monte Carlo (VMC); Harmonic oscillator; quantum mechanical systems; Statistical physics. AMS Subject Classification: 11k45 1. Introduction A basic problem in quantum mechanics is to find the energy eigenstates of quantum mechanical systems. The simplest such system is the quantum harmonic oscillator in one dimension. The Monte Carlo method is commonly used in physics to simulate complex systems those have 376 Farshid Mehrdoust, Hossein Aminikhah, Mohammad Ghamgosar / TJMCS Vol .3 No.4 (2011) 376 - 381 random nature in statistical physics [1]. 2. Variational Monte Carlo The variational method is a powerful tool to estimate the ground state energy of a quantum system, and some of its low lying excited states, when an exact solution cannot be found.When combined with Monte Carlo methods, it can be used on a wide range of atomic and solid state systems, as discussed in the review article by Foulkes et al [2].
    [Show full text]