Computation of observables in discrete stochastic, possibly conditioned, dynamics over large sparse networks is fundamental for a myriad of applications. The Matrix-Product Belief Propagation method allows a semi-analytical estimation of such observables in tree-like networks with a controlled error that depends on the bond dimension, that is the dimension of the employed square matrices. Its computational cost is linear in the time horizon and the network size for a large family of models with discrete degrees of freedom. Here, a generalization of this method to models with continuous or mixed continuous/discrete degrees of freedom is presented, using a tunable expansion in a Hilbert function basis. For a wide class of models, the computational cost of the method is linear in the network size with a prefactor that depends on the basis size and the bond dimension. The method's efficacy is demonstrated by employing a Fourier basis for a mixed continuous/discrete representation of the Kinetic Ising dynamics with real-valued random couplings, where intermediate "local fields" are treated as continuous. The accuracy of the method is verified via comparison with Monte Carlo simulations. For this model, we calculate time auto-correlations, time evolution of energy and magnetization, and finally we estimate the large deviation function of the magnetization at a given future time.
Computation of observables in discrete stochastic, possibly conditioned, dynamics over large sparse networks is fundamental for a myriad of applications. The Matrix-Product Belief Propagation method allows a semi-analytical estimation of such observables in tree-like networks with a controlled error that depends on the bond dimension, that is the dimension of the employed square matrices. Its computational cost is linear in the time horizon and the network size for a large family of models with discrete degrees of freedom. Here, a generalization of this method to models with continuous or mixed continuous/discrete degrees of freedom is presented, using a tunable expansion in a Hilbert function basis. For a wide class of models, the computational cost of the method is linear in the network size with a prefactor that depends on the basis size and the bond dimension. The method's efficacy is demonstrated by employing a Fourier basis for a mixed continuous/discrete representation of the Kinetic Ising dynamics with real-valued random couplings, where intermediate "local fields" are treated as continuous. The accuracy of the method is verified via comparison with Monte Carlo simulations. For this model, we calculate time auto-correlations, time evolution of energy and magnetization, and finally we estimate the large deviation function of the magnetization at a given future time.
Abstract Stochastic dynamics of classical degrees of freedom, defined on vertices of locally tree-like graphs, can be studied in the framework of the dynamic cavity method which is exact for tree graphs. Such models correspond for example to spin-glass systems, Boolean networks, neural networks, and other technical, biological, and social networks. The central objects in the cavity method are edge messages—conditional probabilities of two vertex variable trajectories. In this paper, we discuss a rather pedagogical derivation for the dynamic cavity method, give a detailed account of the novel matrix product edge message (MPEM) algorithm for the solution of the dynamic cavity equation as introduced in Barthel et al (2018 Phys. Rev . E 97 010104 (R)), and present optimizations and extensions. Matrix product approximations of the edge messages are constructed recursively in an iteration over time. Computation costs and precision can be tuned by controlling the matrix dimensions of the MPEM in truncations. Without truncations, the dynamics is exact. Data for Glauber–Ising dynamics shows a linear growth of computation costs in time. In contrast to Monte Carlo simulations, the approach has a much better error scaling. Hence, it gives for example access to low probability events and decaying observables like temporal correlations. We discuss optimized truncation schemes and an extension that allows to capture models which have a continuum time limit.
We describe and demonstrate an algorithm for the efficient simulation of generic stochastic dynamics of classical degrees of freedom defined on the vertices of a locally tree-like graph. Networks with cycles are treated in the framework of the cavity method. Such models correspond for example to spin-glass systems, Boolean networks, neural networks, or other technological, biological, and social networks. Building upon ideas from quantum many-body theory, the algorithm is based on a matrix product approximation of the so-called edge messages -- conditional probabilities of vertex variable trajectories. The matrix product edge messages (MPEM) are constructed recursively. Computation costs and precision can be tuned by controlling the matrix dimensions of the MPEM in truncations. In contrast to Monte Carlo simulations, the approach has a better error scaling and works for both, single instances as well as the thermodynamic limit. As we demonstrate at the example of Glauber dynamics, due to the absence of cancellation effects, observables with small expectation values can be evaluated reliably, allowing for the study of decay processes and temporal correlations.
Out-of-equilibrium dynamics of non-integrable Hamiltonian many-body quantum systems are characterized by highly entangled wave functions. Near-maximal entanglement arises in systems exhibiting thermalization or pre-thermalization, where the system converges to a steady state with a fixed energy density. Classical simulation of the time dependence of such wave functions requires exponential resources. However, typical computations aim to estimate expectation values of local operators and correlation functions to some expected precision. For thermalizing systems at sufficiently high energy densities, such computations can be done without storing the full wave function by instead simulating the evolution of the local operator, which requires significantly fewer resources. Nonetheless, constructing such resource-efficient classical algorithms remains a challenge for intermediate energy densities, where simulating both the wave function and operator evolution is costly. In this paper, we propose a heuristic approach to accelerate the convergence of Matrix Product State (MPS) simulations of expectation values, applicable across a broad range of energy densities. We estimate the desired observables by rescaling the MPS results at low bond dimensions with a factor that depends on the fidelity of the MPS wave function. Using this technique, we simulated the dynamics of the two-dimensional Transverse-Field Ising Model (TFIM) on a 7×8 grid with periodic boundary conditions, using a maximum bond dimension of χ=4096 on a single A100 GPU, as well as the dynamics of the two-dimensional XY model on grids of size up to 9×9. We compare our TFIM results to similar simulations on a digital quantum processor [R. Haghshenas et al., Nature 653, 56 (2026)], demonstrating excellent agreement and confirming the predictive power of our method.
We introduce and apply an efficient method for the precise simulation of stochastic dynamical processes on locally treelike graphs. Networks with cycles are treated in the framework of the cavity method. Such models correspond, for example, to spin-glass systems, Boolean networks, neural networks, or other technological, biological, and social networks. Building upon ideas from quantum many-body theory, our approach is based on a matrix product approximation of the so-called edge messages-conditional probabilities of vertex variable trajectories. Computation costs and accuracy can be tuned by controlling the matrix dimensions of the matrix product edge messages (MPEM) in truncations. In contrast to Monte Carlo simulations, the algorithm has a better error scaling and works for both single instances as well as the thermodynamic limit. We employ it to examine prototypical nonequilibrium Glauber dynamics in the kinetic Ising model. Because of the absence of cancellation effects, observables with small expectation values can be evaluated accurately, allowing for the study of decay processes and temporal correlations.
Tensor network states such as tree tensor network states (TTNS) or matrix product states (MPS) allow very compact representations of high-dimensional wave functions. Using them efficiently relies on a compact representation of the Hamiltonian, and many methods are limited to the sum-of-products (SOP) form. General potential energy surfaces (PESs) do not have this form. To overcome this issue, a stochastic discrete variable representation (SDVR) is presented. The matrix elements of the potential operator in the node-local basis are estimated stochastically by importance sampling of grid configurations. The estimator of the matrix elements is unbiased, and the PES is only evaluated pointwise. The variance is reduced by representing parts of the potential with a 2-mode representation and only sampling the remainder. Sampling noise biases the eigenvalues. A noise-control scheme mitigates this bias. Final energies are obtained from an unbiased variational Monte Carlo estimator. The method is demonstrated by calculating vibrational states of CH3 (6D) and acetonitrile (12D), as well as the zero-point vibrational energy of uracil (30D). For all three systems, the energies agree with the exact references to sub-wavenumber accuracy.
We propose an analytic approach for the steady-state dynamics of Markov processes on locally tree-like graphs. It is based on time-translation invariant probability distributions for edge trajectories, which we encode in terms of infinite matrix products. For homogeneous ensembles on regular graphs, the distribution is parametrized by a single d× d× r^2 d × d × r 2 tensor, where r r is the number of states per variable, and d d is the matrix-product bond dimension. While the method becomes exact in the large- d d limit, it typically provides highly accurate results even for small bond dimensions d d . The d^2r^2 d 2 r 2 parameters are determined by solving a fixed point equation, for which we provide an efficient belief-propagation procedure. We apply this approach to a variety of models, including Ising-Glauber dynamics with symmetric and asymmetric couplings, as well as the SIS model. Even for small d d , the results are compatible with Monte Carlo estimates and accurately reproduce known exact solutions. The method provides access to precise temporal correlations, which, in some regimes, would be virtually impossible to estimate by sampling.