Gradient Flow Method Research Articles

Distributed optimal coordination algorithm for nonlinear second-order multi-agent systems and its application to vehicle platoon

This paper addresses the problem of distributed optimal coordination over second-order multi-agent systems, where the dynamics are nonlinear and unknown. To solve this problem, a state-based observer is designed to handle the unknown matched and unmatched nonlinear terms. Using this observer and the proportional–integral control concept, a distributed optimal coordination algorithm with fixed control parameters is constructed through the consensus scheme and gradient flow method. With the help of the Routh–Hurwitz stability criterion, the exponential convergence proof is guaranteed under the condition that the nonlinear term and the convex local cost function are all locally Lipschitz. Finally, we present four simulation examples, including the vehicle platoon’s cooperative control, to demonstrate the efficiency of our proposed algorithms.

Impact of out-of-plane deformation on atomic reconstruction in twisted van der Waals bilayers

The effects of out-of-plane deformation on atomic relaxation in twisted van der Waals bilayers are investigated by comprehensive multiscale modeling and simulations. The model integrates the DFT-informed generalized stacking-fault energy for interlayer interaction between layers and the Föppl–von Kármán plate theory for elastic energy in each layer with minimization of the total free energy for atomic relaxation governed by a gradient flow method. Our simulation results elucidate twist-angle dependent moiré pattern, strain field, tortional displacement field and stacking domain structures, in good agreement with recent experimental observations. In particular, we derive the strain soliton solution at a small twist angle and the soliton free elastic solution at a large twist angle for the reconstructed van der Waals bilayer with out-of-plane deformation. These results show that out-of-plane deformation not only modifies the strain soliton width but also induces substantial alterations in the strain field, local rotation, and stacking structures. Our findings reveal the non-neglectable role played by out-of-plane deformation in the atomic relaxation of twisted van der Waals bilayers, particularly at smaller twist angles. The intricate interplay between in-plane atomic relaxation and out-of-plane deformation provides opportunities for strain engineering in twisted van der Waals bilayers.

Time-switching patrolling controller for holonomic and nonholonomic multi-robot systems

In this paper, we propose a distributed controller for holonomic and nonholonomic multi-robot systems to patrol multiple targets. Distributed control is a control method that uses only information observed by the robot itself and gathered from its surrounding without relying on centralized control. In order for the robots to accomplish their patrols, we propose an objective function that changes switching weights depending on the time when the targets are not observed. As a result, each robot determines which targets to observe based on the time weights and the information from surrounding robots to achieve efficient patrols. We design a distributed controller based on the gradient-flow method for both holonomic and nonholonomic systems. Finally, simulation and experimental results show that the proposed controller is useful for the robots to realize patrolling.

On the Einstein–Smoluchowski relation in the framework of generalized statistical mechanics

Anomalous statistical distributions that exhibit asymptotic behavior different from the exponential Boltzmann–Gibbs tail are typical of complex systems constrained by long-range interactions or time-persistent memory effects at the stationary non-equilibrium or meta-equilibrium. In this framework, a nonlinear Smoluchowski equation, which models the system’s time evolution towards its steady state, is obtained using the gradient flow method based on a free-energy potential related to a given generalized entropic form. Comparison of the stationary distribution resulting from the maximization of entropy for a canonical ensemble with the steady state distribution resulting from the Smoluchowski equation gives an Einstein-Smoluchowski-like relation. Despite this relationship between the mobility of particle μ and the diffusion coefficient D retains its original expression: μ=βD, appropriate considerations, physically motivated, force us an interpretation of the parameter β different from the traditional meaning of inverse temperature.

Interactions between several types of cosmic strings

We study the interaction of several types of static straight cosmic strings, including local strings, global strings, and bosonic superconducting strings with and without magnetic currents. First, we evaluate the interaction energy of two widely separated cosmic strings using the point source formalism and show that the most dominant contribution to the interaction energy comes from the excitation of the lightest mediator particles in a underlying theory. The interaction energy at arbitrary separation distances is then analyzed numerically by the gradient flow method. It turns out that an additional scalar field introduced in the bosonic superconducting string becomes an additional source of attraction. For such a bosonic superconducting string, we find that a string with two winding numbers is energetically favorable compared to two strings with a single winding number in a certain parameter region. Our analysis reveals that a phase structure of bosonic superconducting strings is richer than that of local and global strings and that the formation of bound states at intersections of bosonic superconducting strings is favored.

Normalized fractional gradient flow for nonlinear Schrödinger/Gross–Pitaevskii equations

In this paper we are interested in gradient flow type methods for computing the ground state of nonlinear Schrödinger/Gross–Pitaevskii equations. Fractional Normalized Gradient Flow methods, involving fractional derivatives and generalizing the celebrated Normalized Gradient Flow method (see Bao and Du, 2004) are derived and analyzed. Several experiments are proposed to illustrate some convergence properties of the developed algorithms.

A modified and efficient phase field model for the biological transport network

This paper aims to establish the biological transport network based on the phase field model. In order to ensure that the topological shape is formed under the guidance of electrical conductivity, we generate the biological networks with sufficient information based on the distributions of the venation of the leaf represented by the reaction-diffusion model. We modify the original energy of the network generating model by considering the auxin gradient property. By applying the gradient flow method to minimize the modified energy, we derive the Poisson type equation for pressure, the reaction-diffusion type equation for the network conductance, and the Allen-Cahn type equation for the phase field. The proposed model is significant on the investigation of phase transitions by considering the gradient properties on the boundaries. We have innovatively added conductivity and phase-field coupling terms that inhibit perpendicular transport of nutrients, making it easy to generate thin branches from the trunk through this model. In order to obtain the second-order temporal accuracy, we take the Crank–Nicolson method for the governing system. To obtain the second-order spatial accuracy, we discretize the coupling system with the central finite difference method and linearize the nonlinear terms semi-explicitly to form a linear system at each time step. The discrete energy dissipation is provably preserved and we can use a larger time step. We apply the preconditioned conjugate gradient method with the multigrid method as a preconditioner to implement a practical algorithm with only linear algebraic complexity. The proposed algorithm is easy to implement and achieves a fast convergence. Various numerical tests are demonstrated to verify the efficiency, stability, and robustness of the proposed method.

Computing the Action Ground State for the Rotating Nonlinear Schrödinger Equation

We consider the computations of the action ground state for a rotating nonlinear Schr\"odinger equation. It reads as a minimization of the action functional under the Nehari constraint. In the focusing case, we identify an equivalent formulation of the problem which simplifies the constraint. Based on it, we propose a normalized gradient flow method with asymptotic Lagrange multiplier and establish the energy-decaying property. Popular optimization methods are also applied to gain more efficiency. In the defocusing case, we prove that the ground state can be obtained by the unconstrained minimization. Then the direct gradient flow method and unconstrained optimization methods are applied. Numerical experiments show the convergence and accuracy of the proposed methods in both cases, and comparisons on the efficiency are discussed. Finally, the relation between the action and the energy ground states are numerically investigated.

Existence of real time quantum path integrals

Many interesting physical theories have analytic classical actions. We show how Feynman’s path integral may be defined non-perturbatively, for such theories, without a Wick rotation to imaginary time. We start by introducing a class of smooth regulators which render interference integrals absolutely convergent and thus unambiguous. The analyticity of the regulators allows us to use Cauchy’s theorem to deform the integration domain onto a set of relevant, complex “thimbles” (or generalized steepest descent contours) each associated with a classical saddle. The regulator can then be removed to obtain an exact, non-perturbative representation. We show why the usual method of gradient flow, used to identify relevant saddles and steepest descent “thimbles” for finite-dimensional oscillatory integrals, fails in the infinite-dimensional case. For the troublesome high frequency modes, we replace it with a method we call “eigenflow” which we employ to identify the infinite-dimensional, complex “eigenthimble” over which the real time path integral is absolutely convergent. We then bound the path integral over high frequency modes by the corresponding Wiener measure for a free particle. Using the dominated convergence theorem we infer that the interacting path integral defines a good measure. While the real time path integral is more intricate than its Euclidean counterpart, it is superior in several respects. It seems particularly well-suited to theories such as quantum gravity where the classical theory is well developed but the Euclidean path integral does not exist.

Glueball spectroscopy in lattice QCD using gradient flow

Removing ultraviolet noise from the gauge fields is necessary for glueball spectroscopy in lattice QCD. It is known that the Yang-Mills gradient flow method is an alternative approach instead of link smearing or link fuzzing in various aspects. In this work we study the application of the gradient flow technique to the construction of the extended glueball operators. We examine a simple application of the gradient flow method, which has some problems in glueball mass calculations at large flow time because of its nature of diffusion in space-time. To avoid this problem, the spatial links are evolved by the ``spatial gradient flow'', that is defined to restrict the diffusion to spatial directions only. We test the spatial gradient flow in calculations of glueball two-point functions and Wilson loops as a new smearing method, and then discuss its efficiency in comparison with the original gradient flow method and the conventional method. Furthermore, to demonstrate the feasibility of our proposed method, we determine the masses of the three lowest-lying glueball states, corresponding to the $0^{++}$, $2^{++}$ and $0^{-+}$ glueballs, in the continuum limit in the pure Yang-Mills theory.

Construction of a two-phase flow with singular energy by gradient flow methods

We prove the existence of weak solutions to a system of two diffusion equations that are coupled by a pointwise volume constraint. The time evolution is given by gradient dynamics for a free energy functional. Our primary example is a model for the demixing of polymers, the corresponding energy is the one of Flory, Huggins and de Gennes. Due to the non-locality in the equations, the dynamics considered here is qualitatively different from the one found in the formally related Cahn-Hilliard equations.Our angle of attack is from the theory of optimal mass transport, that is, we consider the evolution equations for the two components as two gradient flows in the Wasserstein distance with one joint energy functional that has the volume constraint built in. The main difference to our previous work [6] is the nonlinearity of the energy density in the gradient part, which becomes singular at the interface between pure and mixed phases, and thus requires a more sophisticated analysis.

Preliminary results on deconfinement phase transition in full QCD

We explore the profiles of the flux tube connecting a quark and an antiquark in hightemperature SU(3) lattice gauge theory in close vicinity to the critical temperature of thephase transition. In this work, we consider the more realistic case of the flux tube withdynamical quarks, extending the previous study to SU(3) gauge group and making use ofthe Gradient flow method in smoothing procedure for noise reduction. The profiles of thechromoelectric and chromomagnetic field strengths in the flux tube have been measuredfrom Polyakov loop-plaquette correlations using the highly improved staggered quark(HISQ) action on a lattice with temporal extent.

Computing the least action ground state of the nonlinear Schrödinger equation by a normalized gradient flow

In this paper, we generalize the normalized gradient flow method which was first applied to computing the least energy ground state to compute the least action ground state. A continuous normalized gradient flow (CNGF) will be presented and the action diminishing property will be proved to provide a mathematical justification of the gradient flow with discrete normalization (GFDN). Then we use backward-forward Euler method to further discretize the GFDN in time which leads to the GFDN-BF scheme. It is shown that the GFDN-BF scheme preserves the positivity and diminishes the action unconditionally. We compare it with other three schemes which are modified from corresponding ones designed for the least energy ground state and the numerical results show that the GFDN-BF scheme performs much better than the others in accuracy, efficiency and robustness for large time steps. Extensive numerical results of least action ground states for several types of potentials are provided. We also use our numerical results to verify some existing results and lead to some conjectures.

Quantum nucleation of topological solitons

The chiral soliton lattice is an array of topological solitons realized as ground states of QCD at finite density under strong magnetic fields or rapid rotation, and chiral magnets with an easy-plane anisotropy. In such cases, topological solitons have negative energy due to topological terms originating from the chiral magnetic or vortical effect and the Dzyaloshinskii-Moriya interaction, respectively. We study quantum nucleation of topological solitons in the vacuum through quantum tunneling in 2 + 1 and 3 + 1 dimensions, by using a complex ϕ4 (or the axion) model with a topological term proportional to an external field, which is a simplification of low-energy theories of the above systems. In 2 + 1 dimensions, a pair of a vortex and an anti-vortex is connected by a linear soliton, while in 3 + 1 dimensions, a vortex is string-like, a soliton is wall-like, and a disk of a soliton wall is bounded by a string loop. Since the tension of solitons can be effectively negative due to the topological term, such a composite configuration of a finite size is created by quantum tunneling and subsequently grows rapidly. We estimate the nucleation probability analytically in the thin-defect approximation and fully calculate it numerically using the relaxation (gradient flow) method. The nucleation probability is maximized when the direction of the soliton is perpendicular to the external field. We find a good agreement between the thin-defect approximation and direct numerical simulation in 2 + 1 dimensions if we read the vortex tension from the numerics, while we find a difference between them at short distances interpreted as a remnant energy in 3 + 1 dimensions.

Quasi-self-dual Skyrme model

A modification of the Skyrme model has been recently proposed, which admits an exact self-dual sector by the introduction of six scalar fields assembled in a symmetric, positive, and invertible $3\ifmmode\times\else\texttimes\fi{}3$ matrix $h$. In this paper we study soft manners of breaking the self-duality of that model. The crucial observation is that the self-duality equations impose distinct conditions on the three eigenvalues of $h$, and on the three fields lying in the orthogonal matrix that diagonalizes $h$. We keep the self-duality equations for the latter, and break those equations associated to the eigenvalues. We perform the breaking by the addition of kinetic and potential terms for the $h$ fields, and construct numerical solutions using the gradient flow method to minimize the static energy. It is also shown that the addition of just a potential term proportional to the determinant of $h$ leads to a model with an exact self-dual sector and with self-duality equations differing from the original ones by just an additional coupling constant.

Numerical Simulations on Nonlinear Quantum Graphs with the GraFiDi Library

Nonlinear quantum graphs are metric graphs equipped with a nonlinear Schrödinger equation. Whereas in the last ten years they have known considerable developments on the theoretical side, their study from the numerical point of view remains in its early stages. The goal of this paper is to present the Grafidi library, a Python library which has been developed with the numerical simulation of nonlinear Schrödinger equations on graphs in mind. We will show how, with the help of the Grafidi library, one can implement the popular normalized gradient flow and nonlinear conjugate gradient flow methods to compute ground states of a nonlinear quantum graph. We will also simulate the dynamics of the nonlinear Schrödinger equation with a Crank-Nicolson relaxation scheme and a Strang splitting scheme. Finally, in a series of numerical experiments on various types of graphs, we will compare the outcome of our numerical calculations for ground states with the existing theoretical results, thereby illustrating the versatility and efficiency of our implementations in the framework of the Grafidi library.

Preconditioned Pseudo-Spectral Gradient Flow for Computing the Steady-State of Space Fractional Cahn-Allen Equations With Variable Coefficients

The aim of this paper is to propose some efficient and accurate numerical methods to compute the steady-state of variable coefficients space fractional Cahn-Allen equations. The approach combines an adaptive time stepping semi-implicit gradient flow method to minimize the fractional energy functional and pseudo-spectral approximation schemes. Based on the use of a preconditioned GMRES, the space fractional Cahn-Allen equation is then solved efficiently. The full methodology is supported by the numerical solution of a one-dimensional problem.

Modified Gradient Flow Method for Solving One-Dimensional Optimal Control Problem Governed by Linear Equality Constraint

This study presents a computational technique developed for solving linearly constraint optimal control problems using the Gradient Flow Method. This proposed method, called the Modified Gradient Flow Method (MGFM), is based on the continuous gradient flow reformulation of constrained optimization problem with three-level implicit time discretization scheme. The three-level splitting parameters for the discretization of the gradient flow equations are such that the sum of the parameters equal to one (\theta1 + \theta2 +\theta3=1). The Linear and quadratic convergence of the scheme were analyzed and were shown to have first order scheme when each parameter exist in the domain [0, 1] and second order when the third parameter equal to one. Numerical experiments were carried out and the results showed that the approach is very effective for handling this class of constrained optimal control problems. It also compared favorably with the analytical solutions and performed better than the existing schemes in terms of convergence and accuracy

Computing the Least Action Ground State of the Nonlinear Schrödinger Equation By A Normalized Gradient Flow

In this paper, we generalize the normalized gradient flow method which is first applied to computing the least energy ground state to compute the least action ground state. A continuous normalized gradient flow (CNGF) will be presented and the action decreasing property will be proved to provide a mathematical justification of the gradient flow with discrete normalization (GFDN). Then we use backward-forward Euler method to further discretize the GFDN in time which leads to the GFDN-BF scheme. We compare it with other three schemes which are modified from corresponding ones designed for the least energy ground state and the numerical results show that the GFDN-BF scheme performs much better than the others in accuracy, efficiency and robustness for large time steps. Extensive numerical results of least action ground states for several types of potentials are provided. We also use our numerical results to verify some existing results and lead to some conjectures.

Online Optimization of Dynamical Systems With Deep Learning Perception

This paper considers the problem of controlling a dynamical system when the state cannot be directly measured and the control performance metrics are unknown or only partially known. In particular, we focus on the design of data-driven controllers to regulate a dynamical system to the solution of a constrained convex optimization problem where: i) the state must be estimated from nonlinear and possibly high-dimensional data; and ii) the cost of the optimization problem – which models control objectives associated with inputs and states of the system – is not available and must be learned from data. We propose a data-driven feedback controller that is based on adaptations of a projected gradient-flow method; the controller includes neural networks as integral components for the estimation of the unknown functions. Leveraging stability theory for perturbed systems, we derive sufficient conditions to guarantee exponential input-to-state stability (ISS) of the control loop. In particular, we show that the interconnected system is ISS with respect to the approximation errors of the neural network and unknown disturbances affecting the system. The transient bounds combine the universal approximation property of deep neural networks with the ISS characterization. Illustrative numerical results are presented in the context of robotics and control of epidemics.