
We propose a novel class of generalized complex It & ocirc;-type Hermite polynomials and examine their algebraic and operational structure. Connections with classical special functions and polynomials are established, demonstrating links to hypergeometric-type functions. A quasi-monomial structure approach is employed by introducing appropriate raising and lowering operators, leading to operational formulas and second-order differential relations. Rodrigues-type and Burchnall-type operator images are constructed, offering concise expressions and generating relations. Further supplementary results, including generating functions and functional expansions, are derived. Furthermore, the paper demonstrates the applied utility of these generalized polynomials by establishing explicit solutions to one-and two-dimensional heat-type partial differential equations. Graphical representations of the generalized complex Hermite polynomials are also included. The results extend the framework of complex Hermite polynomials and provide a flexible toolset for operational and applied analysis in mathematical physics.
We consider the problem of fitting points on a smooth planar curve or on a smooth surface in the space. To solve this problem, we use inverse multiquadric radial basis functions to construct smooth approximations of the unknown parametrization of the curve or surface from which the data have been sampled. The proposed strategy is based on two main elements, i.e. a specific choice of centers distribution and an automatic selection of the shape parameter. Numerical experiments show the effectiveness and the accuracy of this heuristic technique. An application to real data captured through video recording of a swimmer’s wrist during the entire stroke cycle is considered.
Simulating complex nonlinear physical phenomena typically entails the coupling of multiple physical fields at interfaces. Discontinuities across the interface and nonlinearities pose significant challenges for traditional numerical methods. Neural network (NN) methods combined with piecewise schemes have been developed to resolve the difficulties owing to their meshless feature and powerful nonlinear approximation capability. Deep learning methods generally achieve relative L2 errors on the order of 10−3 or 10−4 at best. To overcome this limitation in accuracy, we propose a piecewise randomized shallow neural network (PRSNN), which is based on a Gauss-Newton optimization framework. A key innovation of our method is an adaptive strategy that dynamically decays the penalty weights based on the relative decrease rate of the subdomain losses. Initially, this strategy constructs a well-conditioned quadratic model to stabilize early iterations. As training progresses, it gradually transitions to the original problem, ensuring asymptotic convergence. We refer to the new method as the piecewise RSNN with decaying penalties (DP-PRSNN). Compared with most standard piecewise models, the DP-PRSNN exhibits superior accuracy, iterative stability, and robustness to initial guesses. The merits have been shown through a wide range of nonlinear interface problems, including those with moving interfaces, multiple interfaces, and nonlinear interface exchange conditions.