Seminar khoa học của Nghiên cứu sinh ngành Khoa học tính toán

Ngày 06/11/2026, Viện IAST tổ chức buổi seminar khoa học của Nghiên cứu sinh ngành Khoa học tính toán tại Phòng họp khán đài A, cơ sở Tân Hưng với nội dung chi tiết như sau:

TÓM TẮT CÁC BÁO CÁO

1. Application of Machine Learning for Predicting the Band Gaps of Materials

Báo cáo viên: Huỳnh Thị Phương Thúy 

Abstract

The band gap is an important property that influences the electronic characteristics and potential applications of materials. However, determining this property using conventional computational methods often requires considerable time and computational resources. Machine Learning has therefore attracted increasing interest in materials research, as data-driven approaches can narrow down the number of materials that need to be investigated using conventional methods, thereby reducing the time and cost of developing new materials. This study applies various machine-learning algorithms to predict the band gaps of different materials. Among the models tested, the Random Forest regression model achieved the highest coefficient of determination (R²) at 0.6435, with a mean absolute error (MAE) of 0.5752 eV and a root mean squared error (RMSE) of 0.9071 eV.

2. [Journal Club] Mass Spectrometric Footprinting and DFT Calculations Explain the Effect of Curcumin on the Aggregation of Amyloid β 42 (Aβ42)

Báo cáo viên: Đỗ Anh Tuấn 

Tham khảo công trình của George Mathai, Michael Gross et al. (Journal of the American Chemical Society, 2026, 148 (34))

Abstract

Curcumin, a polyphenolic natural product, binds to Aβ42, a protein involved in Alzheimer’s Disease, to inhibit self aggregation in vitro. Molecular understanding of the binding is important in designing new molecules to inhibit Aβ42 aggregation. OH radical footprinting of Aβ42 in the presence and absence of curcumin was performed by using fast photochemical oxidation of proteins (FPOP), followed by mass spectral analysis, to monitor changes in Aβ42 protein dynamics in the presence of curcumin. The results reveal that curcumin binding reduces oxidative footprinting at the N-terminus and middle region, indicating binding at these sites. Kinetic modeling of the percentage modification shows a significant change in the aggregation of Aβ42 in the presence of curcumin, whereby one aggregate population nearly disappears. The [Aβ42:curcumin] binding complexes were also explored using density functional theory (DFT), optimizating the starting structures at the B3LYP/6−31G(d) level for both the N-terminal/middle and C-terminal/middle regions. The structure of the lowest-energy complex has curcumin binding at the N-terminal plus middle regions, and this was validated by reoptimization using 6−311G(d,p) and applying Grimme’s dispersion function (GD3). This structure was further confirmed by reoptimization in aqueous media using the Conductor-like Polarizable Continuum Model (CPCM). The new structure not only accounts for the observed changes in •OH footprinting of the N-terminal, middle, and C-terminal regions but also forecasts how this approach can provide insight into protein aggregation. A molecular-level understanding of the binding site of curcumin with Aβ42 should be helpful in the design of new aggregation modulators.

3. Wavelet-Based Numerical Methods for Fractional Differential Equations

Báo cáo viên: Ngô Thị Bích Hoa 

Abstract

Fractional differential equations (FDEs) describe processes with memory and hereditary effects, with applications in population dynamics, biochemical reactions, and infectious disease modeling. FDEs with delay additionally incorporate delayed feedback. Analytical solutions are generally difficult to obtain, motivating the development of numerical methods. Among these approaches, wavelet expansions provide flexible approximations through local basis functions. This seminar presents a spectral collocation method based on fractional-order generalized Legendre wavelets for nonlinear fractional delay systems. Regularized incomplete beta functions yield exact expressions for the Riemann–Liouville fractional integrals of the basis functions, retaining memory contributions beyond their local supports. These expressions are used to reconstruct approximate solutions from wavelet expansions of the Caputo derivatives while satisfying the initial conditions exactly. Incorporating prescribed histories and delayed states reduces the problem to a finite nonlinear algebraic system. Numerical benchmarks and applications to population and biomedical models illustrate the method’s accuracy and the influence of approximation parameters.

Keywords: Fractional differential equations; delay systems; Legendre wavelets; fractional integration; spectral collocation.

4. Wake transitions of the flow around two side-by-side elliptic cylinders

Báo cáo viên: Phan Thị Diệu Thúy 

Abstract

The wake transitions of the flow past two side-by-side elliptical cylinders were numerically investigated using the lattice Boltzmann method at Reynolds numbers (Re) of 40, 100, and 150, with various spacing ratios (L/D, where L represents the distance between the cylinder centers and D is their diameter) and aspect ratios (AR). This study elucidated the effects of Re, AR, and L/D on the flow characteristics, including wake structures, hydrodynamic coefficients, Strouhal number (St), and the spectral energy of the flow exerted on both cylinders. Six distinct flow patterns are observed in the AR-L/D space, such as steady shear layers, anti-phase synchronized streets, in-phase synchronized streets, single bluff body, flip-flopping, and chaotic flow. These patterns were characterized through detailed analyses, including vorticity contour plots, time histories of drag and lift coefficients, power spectral density, and proper orthogonal decomposition of vorticity fluctuations into deterministic spatial structures. Additionally, flow pattern maps and diagrams of the time-averaged pressure coefficients on the surface of the cylinders were provided to assess the influence of Re, AR, and L/D on the flow behavior. The hydrodynamic coefficients of both cylinders showed near-identical trends with significant variations depending on L/D and AR. When AR is small, the time-averaged drag coefficient and the root-mean-squared lift coefficient of both cylinders are found to be substantially higher than those of an isolated elliptical cylinder. Furthermore, a notable increase in the time-averaged lift coefficient was observed when L/D was small, attributed to the repulsive forces between the cylinders. At higher Reynolds numbers (Re = 100 & 150), substantial differences in St emerge, particularly for smaller AR values, despite the cylinders being in a side-by-side configuration.

5. Estimation-based direct simulation on nodal networks (EDISONN) and higher-order shear deformation theory

Báo cáo viên: Vũ Đông Dương 

Abstract

This report presents the Estimation-based Direct Simulation on Nodal Networks (EDISONN) method combined with the seven-degree-of-freedom Higher-order Shear Deformation Theory (7-DOF HSDT) for the analysis of plate structures. Unlike conventional numerical methods such as the Finite Element Method (FEM), Isogeometric Analysis (IGA), and Meshfree methods (MKM), which rely on interpolation shape functions, element meshes, and Gaussian numerical integration, EDISONN is built upon three fundamental principles: no element meshing, no interpolation shape functions, and no numerical integration. As a result, the method provides a simplified discretization procedure, reduces computational cost, and overcomes several inherent limitations of existing methods. From a theoretical standpoint, the computational domain is discretized using an arbitrarily distributed set of nodes, which are linked through Delaunay triangulation solely for the purpose of establishing the neighborhood structure among nodes. The gradient at each node is estimated directly from the values of its neighboring nodes through a least-squares operator combined with a centroid-based coordinate shifting technique to ensure numerical stability. The integrals in the weak form of the governing equations are evaluated using a nodal quadrature rule with weights computed from the equivalent areas of Delaunay triangles, ensuring the conservation of the computational domain. To balance the nodal contributions during the assembly of the mass matrix, an implicit shape function assumption is proposed, which simultaneously satisfies the Kronecker-delta property and thereby allows essential boundary conditions to be imposed directly, as in the Finite Element Method. The combination of EDISONN with the seven-degree-of-freedom HSDT is a natural fit, since the displacement field of this theory contains only first-order derivatives, which is fully compatible with the first-order gradient operator of EDISONN. Moreover, this combination eliminates the need for the shear correction factor required in first-order shear deformation theory.

6. A stochastic Chebyshev moving Kriging meshfree framework for the static bending of functionally graded plates with random material properties

Báo cáo viên: Trần Quốc Hưng 

Abstract

Real functionally graded material (FGM) plates are not perfectly uniform. Manufacturing makes Young's modulus vary randomly from point to point, so the plate deflection is also random. This work presents a stochastic meshfree framework to predict this variation for the static bending of FGM plates. The framework has four parts. First, the plate is modelled with the Chebyshev moving Kriging (CMK) meshfree method and a C⁰- higher-order shear deformation theory (C⁰-HSDT). Second, the random field of Young's modulus is represented by a Karhunen–Loève (KL) expansion, solved on the same CMK shape functions. Third, a sparse polynomial chaos expansion (PCE), built with least angle regression (LARS) and leave-one-out (LOO) cross-validation, serves as a surrogate model. This surrogate model is verified against direct Monte Carlo simulation using the full-order solver. Fourth, Monte Carlo simulation on this surrogate gives the statistical moments (mean, standard deviation and skewness), the probability density function, the Sobol’ sensitivity indices, the failure probability (P_f) and the reliability index (β) of the deflection.