Number Of Variational Parameters Research Articles

Natural Gradient Variational Bayes Without Fisher Matrix Analytic Calculation and Its Inversion

This article introduces a method for efficiently approximating the inverse of the Fisher information matrix, a crucial step in achieving effective variational Bayes inference. A notable aspect of our approach is the avoidance of analytically computing the Fisher information matrix and its explicit inversion. Instead, we introduce an iterative procedure for generating a sequence of matrices that converge to the inverse of Fisher information. The natural gradient variational Bayes algorithm without analytic expression of the Fisher matrix and its inversion is provably convergent and achieves a convergence rate of order O ( log s / s ) , with s the number of iterations. We also obtain a central limit theorem for the iterates. Implementation of our method does not require storage of large matrices, and achieves a linear complexity in the number of variational parameters. Our algorithm exhibits versatility, making it applicable across a diverse array of variational Bayes domains, including Gaussian approximation and normalizing flow Variational Bayes. We offer a range of numerical examples to demonstrate the efficiency and reliability of the proposed variational Bayes method. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Adaptive projected variational quantum dynamics

We propose an adaptive quantum algorithm to prepare accurate variational time evolved wave functions. The method is based on the projected variational quantum dynamics (pVQD) algorithm, that performs a global optimization with linear scaling in the number of variational parameters. Instead of fixing a variational ansatz at the beginning of the simulation, the circuit is grown systematically during the time evolution. Moreover, the adaptive step does not require auxiliary qubits and the gate search can be performed in parallel on different quantum devices. We apply the algorithm, named adaptive pVQD, to the simulation of driven spin models and fermionic systems, where it shows an advantage when compared to both Trotterized circuits and nonadaptive variational methods. Finally, we use the shallower circuits prepared using the adaptive pVQD algorithm to obtain more accurate measurements of physical properties of quantum systems on hardware. Published by the American Physical Society 2024

Exponential type orbitals with hyperbolic cosine function basis sets for isoelectronic series of the atoms Be to Ne

Abstract Exponential type orbital with hyperbolic cosine basis functions, proposed recently for Hartree–Fock–Roothaan calculations of neutral atoms, are studied in detail for the calculations of isoelectronic series of atoms from Be to Ne. Calculations are performed for the neutral and the first 20 cationic members of the isoelectronic series of each atom in its ground state. Three of the most popular exponential type orbitals (Slater type functions, B functions and ψ (α) functions with α = 2) are combined with modified hyperbolic cosine function cosh(βr + γ) to improve the basis function quality within the minimal basis sets framework. Performances of the basis functions are compared with each other by using the same number of variational parameters in them. The obtained results are also compared with numerical Hartree–Fock and extended Slater type basis set results. The presented accuracy of the minimal basis descriptions of atomic systems supports the usage of these unconventional basis functions in electronic structure and property calculations.

Efficient quantum state tomography with convolutional neural networks

Modern day quantum simulators can prepare a wide variety of quantum states but the accurate estimation of observables from tomographic measurement data often poses a challenge. We tackle this problem by developing a quantum state tomography scheme which relies on approximating the probability distribution over the outcomes of an informationally complete measurement in a variational manifold represented by a convolutional neural network. We show an excellent representability of prototypical ground- and steady states with this ansatz using a number of variational parameters that scales polynomially in system size. This compressed representation allows us to reconstruct states with high classical fidelities outperforming standard methods such as maximum likelihood estimation. Furthermore, it achieves a reduction of the estimation error of observables by up to an order of magnitude compared to their direct estimation from experimental data.

Twisted hybrid algorithms for combinatorial optimization

Proposed hybrid algorithms encode a combinatorial cost function into a problem Hamiltonian and optimize its energy by varying over a set of states with low circuit complexity. Classical processing is typically only used for the choice of variational parameters following gradient descent. As a consequence, these approaches are limited by the descriptive power of the associated states. We argue that for certain combinatorial optimization problems, such algorithms can be hybridized further, thus harnessing the power of efficient non-local classical processing. Specifically, we consider combining a quantum variational ansatz with a greedy classical post-processing procedure for the MaxCut-problem on three-regular graphs. We show that the average cut-size produced by this method can be quantified in terms of the energy of a modified problem Hamiltonian. This motivates the consideration of an improved algorithm which variationally optimizes the energy of the modified Hamiltonian. We call this a twisted hybrid algorithm since the additional classical processing step is combined with a different choice of variational parameters. We exemplify the viability of this method using the quantum approximate optimization algorithm (QAOA), giving analytic lower bounds on the expected approximation ratios achieved by twisted QAOA. We observe that for levels p = 1, …, 5, these lower bounds are comparable to the known lower bounds on QAOA at level p + 1 for high-girth graphs. This suggests that using twisted QAOA can reduce the circuit depth by 4 and the number of variational parameters by 2.

Automatic depth optimization for a quantum approximate optimization algorithm

Quantum approximate optimization algorithm (QAOA) is a hybrid algorithm whose control parameters are classically optimized. In addition to the variational parameters, the right choice of hyperparameter is crucial for improving the performance of any optimization model. Control depth, or the number of variational parameters, is considered as the most important hyperparameter for QAOA. In this paper we investigate the control depth selection with an automatic algorithm based on proximal gradient descent. The performances of the automatic algorithm are demonstrated on seven-node and ten-node max-cut problems, which show that the control depth can be significantly reduced during the iteration whereas achieving an sufficient level of optimization accuracy. With theoretical convergence guarantee, the proposed algorithm can be used as an efficient tool for choosing the appropriate control depth as a replacement of random search or empirical rules. Moreover, the reduction of control depth will induce a significant reduction in the number of quantum gates in circuit, which improves the applicability of QAOA on noisy intermediate-scale quantum devices.

Polynomial scaling of the quantum approximate optimization algorithm for ground-state preparation of the fully connected p -spin ferromagnet in a transverse field

We show that the quantum approximate optimization algorithm (QAOA) can construct, with polynomially scaling resources, the ground state of the fully connected $p$-spin Ising ferromagnet, a problem that notoriously poses severe difficulties to a vanilla quantum annealing (QA) approach due to the exponentially small gaps encountered at first-order phase transition for $p\ensuremath{\ge}3$. For a target ground state at arbitrary transverse field, we find that an appropriate QAOA parameter initialization is necessary to achieve good performance of the algorithm when the number of variational parameters $2P$ is much smaller than the system size $N$ because of the large number of suboptimal local minima. Instead, when $P$ exceeds a critical value ${P}_{N}^{*}\ensuremath{\propto}N$, the structure of the parameter space simplifies, as all minima become degenerate. This allows achieving the ground state with perfect fidelity with a number of parameters scaling extensively with $N$ and with resources scaling polynomially with $N$.

Compressing deep neural networks by matrix product operators

A deep neural network is a parametrization of a multilayer mapping of signals in terms of many alternatively arranged linear and nonlinear transformations. The linear transformations, which are generally used in the fully connected as well as convolutional layers, contain most of the variational parameters that are trained and stored. Compressing a deep neural network to reduce its number of variational parameters but not its prediction power is an important but challenging problem toward the establishment of an optimized scheme in training efficiently these parameters and in lowering the risk of overfitting. Here we show that this problem can be effectively solved by representing linear transformations with matrix product operators (MPOs), which is a tensor network originally proposed in physics to characterize the short-range entanglement in one-dimensional quantum states. We have tested this approach in five typical neural networks, including FC2, LeNet-5, VGG, ResNet, and DenseNet on two widely used data sets, namely, MNIST and CIFAR-10, and found that this MPO representation indeed sets up a faithful and efficient mapping between input and output signals, which can keep or even improve the prediction accuracy with a dramatically reduced number of parameters. Our method greatly simplifies the representations in deep learning, and opens a possible route toward establishing a framework of modern neural networks which might be simpler and cheaper, but more efficient.

Variational study of U(1) and SU(2) lattice gauge theories with Gaussian states in1+1dimensions

We introduce a method to investigate the static and dynamic properties of both Abelian and non-Abelian lattice gauge models in 1+1 dimensions. Specifically, we identify a set of transformations that disentangle different degrees of freedom, and apply a simple Gaussian variational ansatz to the resulting Hamiltonian. To demonstrate the suitability of the method, we analyze both static and dynamic aspects of string breaking for the U(1) and SU(2) gauge models. We benchmark our results against tensor network simulations and observe excellent agreement, although the number of variational parameters in the Gaussian ansatz is much smaller.

Continuum tensor network field states, path integral representations and spatial symmetries

A natural way to generalize tensor network variational classes to quantum field systems is via a continuous tensor contraction. This approach is first illustrated for the class of quantum field states known as continuous matrix-product states (cMPS). As a simple example of the path-integral representation we show that the state of a dynamically evolving quantum field admits a natural representation as a cMPS. A completeness argument is also provided that shows that all states in Fock space admit a cMPS representation when the number of variational parameters tends to infinity. Beyond this, we obtain a well-behaved field limit of projected entangled-pair states (PEPS) in two dimensions that provide an abstract class of quantum field states with natural symmetries. We demonstrate how symmetries of the physical field state are encoded within the dynamics of an auxiliary field system of one dimension less. In particular, the imposition of Euclidean symmetries on the physical system requires that the auxiliary system involved in the class’ definition must be Lorentz-invariant. The physical field states automatically inherit entropy area laws from the PEPS class, and are fully described by the dissipative dynamics of a lower dimensional virtual field system. Our results lie at the intersection many-body physics, quantum field theory and quantum information theory, and facilitate future exchanges of ideas and insights between these disciplines.

Optimizing large parameter sets in variational quantum Monte Carlo

We present a technique for optimizing hundreds of thousands of variational parameters in variational quantum Monte Carlo. By introducing iterative Krylov subspace solvers and by multiplying by the Hamiltonian and overlap matrices as they are sampled, we remove the need to construct and store these matrices and thus bypass the most expensive steps of the stochastic reconfiguration and linear method optimization techniques. We demonstrate the effectiveness of this approach by using stochastic reconfiguration to optimize a correlator product state wave function with a Pfaffian reference for four example systems. In two examples on the two dimensional Fermionic Hubbard model, we study 16 and 64 site lattices, recovering energies accurate to 1$%$ in the smaller lattice and predicting particle-hole phase separation in the larger. In two examples involving an ab initio Hamiltonian, we investigate the potential energy curve of a symmetrically dissociated $4\ifmmode\times\else\texttimes\fi{}4$ hydrogen lattice as well as the singlet-triplet gap in free base porphin. In the hydrogen system we recover 98$%$ or more of the correlation energy at all geometries, while for porphin we compute the gap in a 24 orbital active space to within 0.02 eV of the exact result. The number of variational parameters in these examples ranges from $4\ifmmode\times\else\texttimes\fi{}{10}^{3}$ to $5\ifmmode\times\else\texttimes\fi{}{10}^{5}$.

Singer polymals. II. Tempering methods

Tempering of Singer polymal basis sets is introduced to reduce the number of variational parameters to be optimized. Two types of tempering are studied and compared: generalized “even tempering” and generalized “Monte Carlo tempering.” Comparisons of total energy are made for various temperings of Singer geminals using the ground state of the helium atom for a model system. Generalized Monte Carlo tempering is found to be superior. The benefit-to-cost ratio is measured for extending the basis size versus extending the number of tempering parameters; for small basis and small numbers of parameters it is more beneficial to add more parameters, but for sufficiently large basis it is more beneficial to extend the basis than to add more parameters. Monte Carlo tempering exploits the benefits of extending the basis without increasing the number of variational parameters.

Parametrization of the orbital mixing coefficients in the AMO method

It is shown that there is a strong correlation between the values of the orbital mixing coefficients that minimize the energy of the singlet ground state of some alternant conjugated molecules in the AMO approximation. This correlation can be described approximately by a linear relation between a simple function of these coefficients and the suitably defined “orbital energy difference”. The relation is exploited by reducing the number of variational parameters in the AMO method.

A Hylleraas functional based perturbative technique to relax the extremely localized molecular orbital wavefunction

A way to reduce the computational cost associated with the study of large molecules exploits the transfer of extremely localized molecular orbitals (ELMOs). Unfortunately, due to the reduction in the number of variational parameters, the ELMO description is not as accurate as the Hartree-Fock one, although it is qualitatively correct in most of the cases. Therefore, in order to overcome this significant drawback, we propose a perturbative ELMO approach exploiting the Hylleraas functional. Preliminary tests have been performed and the results are promising for future applications to large systems.

Extremely localized molecular orbitals: theory and applications

Orbitals that are extremely localized on molecular fragments represent a powerful tool for a number of purposes: to cite a few examples, they allow to reduce strongly the complexity of calculations on large systems and are easily transferable from one molecule to another, providing a suitable and efficient way to build up the electronic structure of large molecules. Recently, we have developed efficient algorithms to determine extremely localized molecular orbitals (ELMOs), which will be reviewed in this paper. As a rigorous localization is strictly connected to a reduction in the number of variational parameters, which reflects into an increased value of the associated energy with respect to the Hartree Fock value, we have developed a number of strategies to relax the wavefunction built up using transferred localized orbitals. The extreme localization has also been exploited in connection with the “Divide and Conquer” technique to determine the electron densities of large polypeptides assembled from orbitals computed on small model molecules. Moreover, we will discuss the recent application of the ELMOs in the framework of the hybrid QM/MM methods to describe the frontier region. We will also show that the ELMOs can be used to extract chemical interpretations from numerical results. A variety of applications will be presented.

A New Antisymmetrized Molecular Dynamics Approach to Few-Nucleon Systems

A new method in the study of antisymmetrized molecular dynamics (AMD) is proposed. This method consists of a Jacobi-coordinate-basis AMD with the generator coordinate method (GCM). A simple example is presented for the case of in few-nucleon systems with the Volkov potential. Bound states for 2N, 3N, and 4N systems are compared with those obtained using the “primary AMD” and some other methods. We find this to be a promising method for investigating light nuclear systems. During the past decade, since the primary antisymmetrized molecular dynamics (AMD) was proposed, 1)–3) it has been successfully extended to systematic studies of neutron-rich nuclei, 4)–6) as well as excited states of nuclei. 7) Further developments of the AMD method have been realized by introducing several conventional ideas of conventional nuclear structure physics. 8)–19) The physical approaches mentioned above were formulated by making improvements of the AMD in Refs. 4)–7); for instance, a variational calculation was carried out after application of the spin parity projection (VAP) in order to investigate excited states 15) and the existence of exotic clusters in excited states of neutronrich nuclei. 16)–19) Another improvement realized through use of the superposition of selected Slater determinants was made in the AMD triple-S model, which can quantitatively describe properties of light nuclei. 10),11) These improvements were made in methods treating medium light nuclei in which the wave function is expressed in terms of a superposition of Slater determinants. However, this method has not yet been successfully applied to few-body problems. In order to allow for the application of the AMD method to few-body and very light nuclei, we present one modified AMD method which was developed using the Jacobi-coordinate-basis AMD plus the GCM technique. 20),21) One of the aims of this work is to introduce the N-body correlations in a “single Slater determinant”. The second is to remove the ambiguity of the zero-point oscillation energy. The third is to reduce the number of variational parameters in the AMD effectively with the GCM technique; using this technique, a very large number of the variational parameters can be fixed in the eigenvalue problem. Let us start from the trial wave function primary AMD. The wave function of N -nucleon system Ψ is given by

A novel approach to relax extremely localized molecular orbitals: the extremely localized molecular orbital?valence bond method

Orbitals rigorously defined on molecular fragments, like the extremely localized molecular orbitals, represent a possible way to build up the electronic structure of large molecules, using molecular fragments assembled on small molecules. Of course a rigorous localization is strictly connected to a reduction in the number of variational parameters, which reflects itself in an increased value of the associated energy. In order to get a more accurate description of the target molecule, we have developed a method which allows a relaxation of the electronic structure based on transferred localized orbitals. The relaxation is realized by means of a valence bond technique, which in turn uses the localized nature of the orbitals to reduce the number of excitations. Applications to the transferability of extremely localized molecular orbitals are presented.

Titrating PolyelectrolytesVariational Calculations and Monte Carlo Simulations

Variational methods are used to calculate structural and thermodynamical properties of a titrating polyelectrolyte in a discrete representation. In the variational treatment, the Coulomb potentials are emulated by harmonic repulsive forces between all monomers; the force constants are used as variational parameters. The accuracy of the variational approach is tested against Monte Carlo data. Excellent agreement is obtained for the end-to-end separation and the apparent dissociation constant for the unscreened Coulomb chain. The short-range screened Coulomb potential is more difficult to handle variationally, and its structural features are less well described, although the thermodynamic properties are predicted with the same accuracy as for the unscreened chain. The number of variational parameters is on the order of N2, where N is the number of monomers, and the computational effort scales like N3. In addition, a simplified variational procedure with only two parameters is pursued, based on a rigid-rod approximation of the polymer. It gives surprisingly good accuracy for certain physical properties.

Generalized dimer states: hierarchies

We present variational calculations for quantum spin-half models based upon modifications to nearest-neighbour dimer states. Using a sequence of projections we can attempt a sort of finite-size scaling, in the number of variational parameters, towards the exact solution. For the simple nearest-neighbour Heisenberg model we achieve an error of only 0.005% in energy for our most accurate calculation, much better than an equivalent Lanczos calculation. With which we compare. Due to the fact that we control the total spin in our wavefunctions, we can evaluate the coherence lengths associated with cyclic-exchange permutations (the spin-half sector) and for spin-spin correlations (the spin-one sector), yielding, for example, a convincing numerical confirmation that the spin-one biquadratic-exchange Hamiltonian has a finite correlation length. We perform a study of the one-dimensional J1-J2 model, trying to predict the onset of the phase transition by searching directly for the divergence of the correlation length. Our calculations predict the phase transition near J2/J1 approximately 1/4, as expected. We also perform a study of the ladder geometry, giving further evidence that infinitesimal coupling between ladders yields an immediate gap in the spectrum. The only complication is that the minimization over our space of variational parameters is subtle: surprisingly complicated and "difficult to find" structure is observed in our low-energy solutions which might make our approach of limited use.

TRANSFER-MATRIX METHOD FOR SOLVING THE SPIN 1/2 ANTIFERROMAGNETIC HEISENBERG CHAIN

Following the discovery of high Tc superconductivity in the copper oxides, there has been a great deal of interest in the RVB wave function proposed by Anderson [1]. As a warm-up exercise we have considered a valence-bond wave function for the one dimensional spin-1/2 Heisenberg chain. The main virtue of our work is to propose a new variational singlet wavefunction which is almost analytically tractable by a transfer-matrix technique. We have obtained the ground state energy for finite as well as infinite chains, in good agreement with exact results. Correlation functions, excited states, and the effects of other interactions (e.g., spin-Peierls) are also accessible within this scheme [2]. Since the ground state of the chain is known to be a singlet (Lieb & Mattis [3]), we write the appropriate wave function as a superposition of valence-bond singlets, [Formula: see text], where | k > is a spin configuration obtained by pairing all spins into singlet pairs, in a way which is common in valence-bond calculations of large molecules. As in that case, each configuration, | k >, can be represented by a Rümer diagram, with directed bonds connecting each pair of spins on the chain. The c k 's are variational co-efficients, the form of which is determined as follows: Each singlet configuration (Rümer diagram) is divided into "zones", a "zone" corresponding to the region between two consecutive sites. Each zone is indexed by its distance from the end of the chain and by the number of bonds crossing it. Our procedure assigns a variational parameter, x ij , to the j th zone, when crossed by i bonds. The resulting wavefunction for an N-site chain is written as [Formula: see text] where m ij(k) equals 1 when zone j is crossed by i bonds and zero otherwise. To make the calculation tractable we reduce the number of variational parameters by disallowing configurations with bonds connecting any two sites separated by more than 2M lattice points. (For simplicity, we have limited ourselves to M=3, but the scheme can be used for any M). With the simple ansatz, matrix elements can be calculated by a transfer-matrix method. To understand the transfer-matrix method note that since only local zone parameters appear in the description of each state | k >, matrix elements and overlaps, [Formula: see text] and < k | k '>, are completely specified by a small number of "local states" associated with each zone. Within a given zone a local state is defined by (i) the number of bonds crossing the zone and (ii), by whether the bonds originate from the initial (| k >) or final (| k '>) state. It is then easy to see that "local states" of consecutive zones are connected by a 15 × 15 transfer matrix (for the case M=3). Furthermore, the overlap matrix element can be written as a product of transfer-matrices associated with each zone of the chain. When calculating matrix elements of the Hamiltonian, an additional matrix, U , must be defined, to represent the particular zone involving the two spins connected by the Heisenberg interaction. The relevant details as well as the comparison with exact results will be given elsewhere. We are planning to ultimately apply this method to the two dimensional case, and hope to include the effects of holes.