为助力师生追踪力学、数学与应用物理学领域前沿动态，提升科研创新能力，拓宽学术交流视野，理学院特邀请国内外知名专家学者莅临我校开展系列学术讲座。现将具体安排通知如下，欢迎各位师生踊跃参加。
报告一：
报告时间：2026 年11月14日 上午9:00
报告地点：新能源楼 317
报告专家：Semenova Olga 高级讲师，圣彼得堡国立大学
报告题目：An Exact Algorithm for the Shortest Path Problem Under Multiple Resource
Constraints
摘要：Let G = ⟨V, A⟩ be a directed graph, let w, τ₁, …, τn be weight functions on its arcs, let T₁, …, Tn be given real numbers, and let s, t be fixed vertices. For a path μ, its w- and τ-weights are defined as the sums of the weights of its arcs (counted with multiplicity). The talk addresses the exact optimization problemw(μ) → min, μ ∈ M(s, t), τᵢ(μ) ≤ Tᵢ, i = 1, …, N, where M(s, t) is the set of all paths from s to t. The algorithm is based on the Bellman– Moore algorithm, in which the classical relaxation step is replaced by Pareto filtering of path evaluation vectors; this substantially reduces the search space while preserving all potentially optimal paths. A sufficient condition for the correct operation of the algorithm is the absence of cycles of nonpositive weight with respect to at least one of the weights τᵢ. Under additional conditions — integer weights and polynomial bounds on the growth of the problem parameters — the running time admits a polynomial bound in the number of vertices n. The talk presents the correctness proof, the complexity analysis, and numerical examples.
报告二：
报告时间：2026 年11月14日 上午9:30
报告地点：新能源楼 317
报告专家：张强 教授，南京大学
报告题目：Stability Analysis and Error Estimate of Adams-Bashforth Discontinuous
Galerkin Methods for Linear Hyperbolic Equations
摘要：The main context in this talk is to systematically investigate the L2-norm stability of Adams-Bashforth discontinuous Galerkin (ABDG) methods for linear hyperbolic equations. The crucial technique is a novel matrix transferring process based on backward differences of numerical solutions defined at the same time position, which can automatically yield an energy equation facilitating theoretical analysis. After establishing the decay properties among backward differences, we derive the L2-norm stability conclusions for ABDG methods up to sixteenth order. Specifically, many commonly-used methods, including the second order method with piecewise linear polynomials, are proved to be stable under a CFL condition. With the help of the above stability results and the generalized Gauss-Radau projections, we can easily obtain the optimal error estimate along a standard analysis line. Some numerical experiments are also given to support our finding. This work is jointed with Jiatao He, Chi-Wang Shu and Yuan Xu.
报告三：
报告时间：2026 年11月14日 上午10:20
报告地点：新能源楼 317
报告专家：孟雄 教授，哈尔滨工业大学
报告题目：Discontinuous Galerkin methods with generalized numerical fluxes for several time dependent convection-dominated PDEs
摘要：In this talk, we first consider the discontinuous Galerkin method using generalized numerical fluxes for linearized KdV equations. We are able to choose a downwind-biased flux in possession of the anti-dissipation property for the convection term to compensate the numerical dissipation of the dispersion term. This is beneficial to obtain a lower growth of the error and to accurately capture the exact solution without phase errors for long time simulations. By establishing relationships of different numerical viscosity coefficients, a uniform stability is shown. Moreover, a suitable numerical initial condition is chosen. By using generalized Gauss-Radau projections, optimal error estimates are derived. Extensions of discontinuous Galerkin methods with generalized fluxes for nonlinear convection-diffusion systems, nonlinear Schrödinger equations and nonlinear hyperbolic conservation laws are also given. Numerical experiments are provided to confirm the theoretical results.
报告四：
报告时间：2026 年11月14日 上午10:50
报告地点：新能源楼 317
报告专家：卢键方 副教授，华南理工大学
报告题目：An oscillation suppressing approach for discontinuous Galerkin methods
摘要：In this talk, we propose a local constraint projection method for discontinuous Galerkin approximations of hyperbolic conservation laws. Their solutions propagate along characteristics and typically satisfy physical bounds, bounded total variation, and entropy conditions, and may develop finite-time discontinuities when characteristics intersect. High-order linear schemes often generate spurious oscillations, undershoots, and overshoots near such discontinuities, potentially producing nonphysical structures or even causing simulation failure. Moreover, among infinitely many weak solutions, only the entropy solution is physically relevant. To address these issues, we develop a parameterfree local constraint projection limiter. The method preserves solution bounds, suppresses numerical oscillations, controls total variation, and yields entropy-satisfying solutions. Theoretical analysis shows that the proposed local constraints are sufficient to ensure bound preservation, bounded total variation, and entropy stability within the semidiscrete framework. Numerical experiments demonstrate the effectiveness of the method and confirm the theoretical results.
报告五：
报告时间：2026 年11月14日 下午14:00
报告地点：新能源楼 317
报告专家：李小舟 副研究员，电子科技大学
报告题目：Extracting More from Numerical Solutions: Convolution-Based Reconstruction for Accuracy Enhancement
摘要：Numerical solutions may contain useful information that is not immediately visible from their pointwise accuracy or regularity. For certain discretizations, structured components of the numerical error retain higher-order information in weak, averaged, or locally combined forms. This raises a natural question: can such information be recovered from an already computed solution, without modifying or rerunning the original solver? This talk introduces convolution-based reconstruction techniques developed within the Smoothness-Increasing Accuracy-Conserving (SIAC) framework. Unlike conventional smoothing filters, these reconstructions use carefully designed kernels to preserve polynomial information while exploiting structures in the underlying discretization error. Under suitable conditions, the resulting representation can be both smoother and more accurate than the original numerical solution. We discuss the main principles behind the reconstruction, including what information it uses, which kernel properties are essential, and how it can be applied non-intrusively to data generated by existing numerical software.
报告六：
报告时间：2026 年11月14日 下午14:30
报告地点：新能源楼 317
报告专家：李寅 副教授，哈尔滨工业大学
报告题目：Multi type mean field reinforcement learning for optimal resource allocation in heterogeneous network
摘要：With the exponential growth in the amount of data transmitted over mobile networks, contemporary 5G communication technologies with the primary goal of improving network performance and quality of service have gained much attention. Efficient resource allocation and interference management are especially critical in large-scale wireless networks. Device-to-device (D2D) communication has become a promising technological tool to address this growing need. However, the limitation of exponentially growing solution space in largescale ultra-dense networks makes it difficult to achieve real-time control with conventional optimization methods. To face this challenge, we propose a novel framework that combines Multi-Agent Reinforcement Learning (MARL) with Mean Field Type Game (MFTG) theory, allowing agents to operate in different action spaces. This approach extends the core principle of mean-field reinforcement learning from a single type to multiple types of interactions, effectively modeling the approximate behavior between various types of devices in heterogeneous D2D networks. Experimental results show that the proposed Multi-Type Mean-Field double deep Q-network (MTMFQ) method outperforms benchmark methods in heterogeneous networks. In addition, the proposed method exhibits good scalability in parameters such as user density, network size and power budget, showing its potential for application in ultra-dense heterogeneous communication network scenarios.
报告七：
报告时间：2026 年11月14日 下午15:00
报告地点：新能源楼 317
报告专家：李佳兴 副教授，哈尔滨理工大学
报告题目：Recursive Filtering for Stochastic Systems under Unreliable Information: A Covariance-Based Framework
摘要：The stochasticity is ubiquitous in practical engineering domains, such as power equipment, autonomous robots and ocean platforms. Typical examples include fluctuations of wind power and photovoltaic systems induced by illumination and wind speed, sensor measurement noises during the navigation of mobile robots, and random ocean current disturbances acting on underwater vehicles. Within the framework of networked sensing and information transmission, system measurement signals are subject to coupled interferences from multiple uncertainties including uniform quantization, network scheduling protocols and deception attacks. This paper is concerned with recursive leastsquares linear estimation algorithms for such complex stochastic systems under unreliable information. To improve communication efficiency under limited bandwidth and alleviate data collisions caused by concurrent transmissions of multiple nodes, a network scheduling mechanism is introduced to regulate the transmission sequence of measurement signals. Meanwhile, the risk of data tampering arising from deception attacks over transmission channels is fully taken into account. To mitigate the adverse effect of data loss triggered by network scheduling on estimation performance, three data compensation strategies, namely zero-input compensation, zero-order hold compensation and prediction-based compensation, are investigated for comparison. With the available measurement information, a covariance-based recursive filter is devised, and a fixed-point smoother is further derived. Several comparative simulation experiments verify the feasibility and superiority of the developed recursive estimation scheme. Furthermore, the influence of deception attack probabilities on the estimation accuracy is quantitatively analyzed.
理学院
2026.11.12