Structures of composite spin operators are analyzed which appear in models of neural networks of the type which Amit et al. have recently investigated. A binary basis of size N${=2}^{M}$ is introduced to study a problem of N quantum-mechanical spin operators. The [Z${(2)]}^{M}$ group structure of the binary basis allows for many decompositions of the [SU(2)${]}^{\mathrm{N}}$ spin algebra. These become useful in studying and solving generalized frustrated Heisenberg as well as Ising models. Using these techniques for quantum-mechanical generalized spin operators, we derive an explicit representation of the partition function of classical statistical-mechanics models, in terms of a series summation over components of collective spin variables.
Structures of composite spin operators are analyzed which appear in models of neural networks of the type which Amit et al. have recently investigated. A binary basis of size N=2M is introduced to study a problem of N quantum-mechanical spin operators. The [Z(2)]M group structure of the binary basis allows for many decompositions of the [SU(2)]N spin algebra. These become useful in studying and solving generalized frustrated Heisenberg as well as Ising models. Using these techniques for quantum-mechanical generalized spin operators, we derive an explicit representation of the partition function of classical statistical-mechanics models, in terms of a series summation over components of collective spin variables.
Frustration is a ubiquitous phenomenon in many-body physics that influences the nature of the system in a profound way with exotic emergent behavior. Despite its long research history, the analytical or numerical investigations on frustrated spin models remain a formidable challenge due to their extensive ground-state degeneracy. In this paper, we propose a unified tensor network theory to numerically solve the frustrated classical spin models on various two-dimensional (2D) lattice geometry with high efficiency. We show that the appropriate encoding of emergent degrees of freedom in each local tensor is of crucial importance in the construction of the infinite tensor network representation of the partition function. The frustrations are thus relieved through the effective interactions between emergent local degrees of freedom. Then the partition function is written as a product of a one-dimensional (1D) transfer operator, whose eigenequation can be solved by the standard algorithm of matrix product states rigorously, and various phase transitions can be accurately determined from the singularities of the entanglement entropy of the 1D quantum correspondence. We demonstrated the power of our general theory by numerically solving 2D fully frustrated XY spin models on the kagome, square, and triangular lattices, giving rise to a variety of thermal phase transitions from infinite-order Brezinskii-Kosterlitz-Thouless transitions, second-order transitions, to first-order phase transitions. Our approach holds the potential application to other types of frustrated classical systems like Heisenberg spin antiferromagnets.
We consider a “generalized Heisenberg model”, with a Hamiltonian given by ℋ︁(D) = –JΣij S(D)j, where the quantities Si are isotropically-interacting D-dimensional classical spins and the summation is restricted to nearest-neighbor pairs of sites <i,j>. Thus when D = 1, 2, and 3, this model reduces to the S = 1/2 Ising, plane rotator, and classical Heisenberg models which are useful for describing phase transitions in, respectively, liquid-gas systems, Bose fluids, and magnetic materials. Unfortunately this Hamiltonian is exactly soluble only in the limit D → ∞, in which case the solution corresponds to the spherical model of Berlin and Kac. For finite D, we must consider approximation procedures in order to obtain any useful information. In this paper we develop a diagrammatic representation, valid for arbitrary D, for successive terms in the high-temperature expansions of the free energy and the spin-spin correlation function—from which we can obtain series expansions for all the thermodynamic properties of the system.
The Heisenberg model, a quantum-mechanical analogue of the Ising model, has a large ground-state degeneracy, due to the symmetry generated by the total spin. This symmetry is also responsible for degeneracies in the rest of the spectrum. We discuss the global structure of the spectrum of Heisenberg models with arbitrary couplings, using group-theoretical methods. The Hilbert space breaks up in blocks characterized by the quantum numbers of the total spin, S and M, and each block is shown to constitute the representation space of an explicitly given irreducible representation of the symmetric group SN, consisting of permutations of the N spins in the system. In the second part of the paper we consider, as a concrete application, the model where each spin is coupled to all the other spins with equal strength. Its partition function is written as a single integral, elucidating its N dependence. This provides a useful framework for studying finite-size effects. We give explicit results for the heat capacity, revealing interesting behavior just around the phase transition.
We review some connections between quantum information and statistical mechanics. We focus on three sets of results for classical spin models. First, we show that the partition function of all classical spin models (including models in different dimensions, different types of many-body interactions, different symmetries, etc) can be mapped to the partition function of a single model. Second, we give efficient quantum algorithms to estimate the partition function of various classical spin models, such as the Ising or the Potts model. The proofs of these two results are based on a mapping from partition functions to quantum states and to quantum circuits, respectively. Finally, we show how classical spin models can be used to describe certain fluctuating lattices appearing in models of discrete quantum gravity.
Machine learning has become increasingly popular as a computational tool in all aspects of science and the private sector, however its application as an additional computational physics tool has only recently began to gain traction. This thesis aims to study the application of supervised and unsupervised neural networks to computational classical statistical physics problems, focusing on what is learned and\nwhether it is a useful tool for these applications. Supervised neural networks are used to distinguish between the different phases of four different models, containing a second-order phase-transition (PT), infinite-order PT, and both in close proximity to one another, as well as a model without a PT but with frustration. Different iterations of the restricted Boltzmann machine (RBM), a type of unsupervised neural\nnetwork, are trained on the one- and two-dimensional Ising model. The neural network is able to differentiate between the different phases only if the PT is clearly discernible from the input configurations. It is too crude a tool to differentiate between two close subsequent PTs. It is concluded that application of neural networks to detect PTs in classical statistical physics models where an intuition for the PT exists, holds no advantage over alternative conventional computational methods. This thesis concludes that restrictions placed on the RBM, such that the RBM has translation\ninvariance, still allow the restricted RBMs to learn the magnetisation really well, while two-spin correlations are learned less well. Surprisingly, the block Gibbs sampling of the restricted RBMs is better behaved than for the unrestricted RBM, as can be explained using an analysis of the trained weight-matrix.
This is the first in a series of papers. It deals with the linked-cluster summation of the perturbation series in quantum mechanics and quantum statistical physics. In particular, a formalism is developed for calculating the matrix element of the operator exp(+sH), where H is the quantum-mechanical Hamiltonian operator. It is found that the matrix element may be written as an ordinary c-number integral of a c-number function over the phase space. This function is found by doing a linked-cluster summation of the usual perturbation series. It is in the form of an exponential of an effective Hamiltonian, Heff, which is a function of the parameter s and the c-number momentum and position coordinates. We have assumed that the Hamiltonian has no spin-dependent terms. The matrix element is written in such a form that it may be used for calculating the transition matrix elements in time-dependent perturbation theory as well as for calculating the canonical partition function ZN in quantum statistical mechanics. The former is obtained by substituting $s=\ensuremath{-}\frac{\mathrm{it}}{\ensuremath{\hbar}}$ and the latter by substituting $s=\ensuremath{-}\ensuremath{\beta}$. The series for Heff depends only on the potential and is independent of the statistics. The effects of the statistics (Bose-Einstein, Fermi-Dirac, or Maxwell-Boltzmann) are all included in a certain factor multiplying exp(sHeff). Thus we have succeeded in separating the statistical and quantum corrections. These latter arise from the noncommutativity of the position and momentum operators. The effective Hamiltonian is an infinite series. A diagrammatic technique is given for evaluating this series. It is an expansion in powers of the coupling constant and can also be expanded in powers of Planck's constant. In general, it is not possible to give a closed-form expression for Heff. But for most practical cases it is sufficient to keep the first few powers of Planck's constant $\ensuremath{\hbar}$, because it is very small. We therefore have an entirely new method of approximation. The statistical effect may always be included exactly. No numerical estimates are made in this paper.