Reinforcement learning (RL) algorithms based on high-dimensional function approximation have achieved tremendous empirical success in large-scale problems with an enormous number of states. However, most analysis of such algorithms gives rise to error bounds that involve either the number of states or the number of features. This paper considers the situation where the function approximation is made either using the kernel method or the two-layer neural network model, in the context of a fitted Q-iteration algorithm with explicit regularization. We establish an $\tilde{O}\left(H^{3}|\mathcal{A}|^{\frac{1}{4}} n^{-\frac{1}{4}}\right)$ bound for the optimal policy with HnHn samples, where HH is the length of each episode and $|\mathcal{A}|$ is the size of action space. Our analysis hinges on analyzing the L² error of the approximated Q-function using nn data points. Even though this result still requires a finite-sized action space, the error bound is independent of the dimensionality of the state space.

This paper is concerned with numerical solutions for the Allen-Cahn equation with standard double well potential and periodic boundary conditions. Surprisingly it is found that using standard numerical discretizations with high precision computational solutions may converge to completely incorrect steady states. This happens for very smooth initial data and state-of-the-art algorithms. We analyze this phenomenon and showcase the resolution of this problem by a new symmetry-preserving filter technique. We develop a new theoretical framework and rigorously prove the convergence to steady states for the filtered solutions.

This paper is concerned with the development of numerical research for the 2D vorticity-stream function formulation and its application in vortex merging at high Reynolds numbers. A novel numerical method for solving the vorticity-stream function formulation of the Navier-Stokes equations at high Reynolds number is presented. We implement the second-order linear scheme by combining the finite difference method and finite volume method with the help of careful treatment of nonlinear terms and splitting techniques, on a staggered-mesh grid system, which typically consists of two steps: prediction and correction. We show in a rigorous fashion that the scheme is uniquely solvable at each time step. A verification algorithm that has the analytical solution is designed to demonstrate the feasibility and effectiveness of the proposed scheme. Furthermore, the proposed scheme is applied to study the vortex merging problem. Ample numerical experiments are performed to show some essential features of the merging of multiple vortices at high Reynolds numbers. Meanwhile, considering the importance of the inversion for the initial position of the vorticity field, we present an iteration algorithm for the reconstruction of the initial position parameters.

In this paper, we establish a generalized population system of natural pinus koraiensis with lactation delay and diffusion term. Firstly, through the eigenvalue analysis, the conditions for local asymptotic stability of the positive equilibrium are derived, and the time delay is taken as bifurcation parameter, the existence conditions of Hopf bifurcation are discussed. Secondly, the model is analyzed qualitatively from the bifurcation point of view. The existence conditions of Turing bifurcation are given. By utilizing the normal form and center manifold theories of partial functional differential equations, the direction of Hopf bifurcation and the stability of bifurcating periodic solutions are studied, and the related formulae are determined. Finally, a nonlinear population model of pinus koraiensis with time delay and diffusion term is established. The corresponding numerical simulations are performed to verify the effects of time delay and diffusion on the stability of system, and the biological explanation is given.

This paper aims at studying the difference between Ritz-Galerkin (R-G) method and deep neural network (DNN) method in solving partial differential equations (PDEs) to better understand deep learning. To this end, we consider solving a particular Poisson problem, where the information of the right-hand side of the equation $f$ is only available at nn sample points, that is, $f$ is known at finite sample points. Through both theoretical and numerical studies, we show that solution of the R-G method converges to a piecewise linear function for the one dimensional problem or functions of lower regularity for high dimensional problems. With the same setting, DNNs however learn a relative smooth solution regardless of the dimension, this is, DNNs implicitly bias towards functions with more low-frequency components among all functions that can fit the equation at available data points. This bias is explained by the recent study of frequency principle. In addition to the similarity between the traditional numerical methods and DNNs in the approximation perspective, our work shows that the implicit bias in the learning process, which is different from traditional numerical methods, could help better understand the characteristics of DNNs.

The backward differentiation formula (BDF) is a popular family of implicit methods for the numerical integration of stiff differential equations. It is well noticed that the stability and convergence of the A-stable BDF1 and BDF2 schemes for parabolic equations can be directly established by using the standard discrete energy analysis. However, such classical analysis seems not directly applicable to the BDF-k with 3 ≤ k ≤ 5. To overcome the difficulty, a powerful analysis tool based on the Nevanlinna-Odeh multiplier technique [Numer. Funct. Anal. Optim., 3:377-423, 1981] was developed by Lubich et al. [IMA J. Numer. Anal., 33:1365-1385, 2013]. In this work, by using the so-called discrete orthogonal convolution kernel technique, we recover the classical energy analysis so that the stability and convergence of the BDF-k with 3 ≤ k ≤ 5 can be established.

A threshold graph can be represented as the binary sequence. In this paper, we present an explicit formula for computing the distance characteristic polynomial of a threshold graph from its binary sequence, and then give a necessary and sufficient condition to characterize two distance cospectral but non-isomorphic threshold graphs. As its applications, we obtain many families of distance cospectral threshold graphs. This provides a negative answer to the problem posed in [22].