Applied Mathematics and Machine Learning
Department of Mathematics and Statistics
Texas Tech University
 | applied_math 2022 fall
| |
|
In this talk, we present our work on structure-preserving machine learning (ML) moment closure models for the radiative transfer equation. Most of the existing ML closure models are not able to guarantee the stability, which directly causes blow up in the long-time simulations. In our work, with carefully designed neural network architectures, the ML closure model can guarantee the stability (or hyperbolicity). Moreover, other mathematical properties, such as physical characteristic speeds, are also discussed. Extensive benchmark tests show the good accuracy, long-time stability, and good generalizability of our ML closure model.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
In the literature, spectral element methods usually refer to finite element methods with high order polynomial basis. The Q^k spectral element method has been a popular high order method for solving second order PDEs, e.g., wave equations, for more than three decades, obtained by continuous finite element method with tenor product polynomial of degree k and with at least (k+1)-point Gauss-Lobatto quadrature. In this talk, I will present some brand new results of this classical scheme, including its accuracy, monotonicity (stability), and examples of using monotonicity to construct high order accurate bound (or positivity) preserving schemes in various applications including the Allen-Cahn equation coupled with an incompressible velocity field, Keller-Segel equation for chemotaxis, nonlinear eigenvalue problem for Gross–Pitaevskii equation, and compressible Navier-Stokes equations.
1) Accuracy: when the least accurate (k+1)-point Gauss-Lobatto quadrature is used, the spectral element method is also a finite difference (FD) scheme, and this FD scheme can sometimes be (k+2)-th order accurate for k>=2. This has been observed in practice but never proven before as rigorous a priori error estimates. We are able to prove it for linear elliptic, wave, parabolic and Schrödinger equations for Dirichlet boundary conditions. For Neumann boundary conditions, (k+2)-th order can be proven if there is no mixed second order derivative. Otherwise, only (k+3/2)-th order can be proven and some order loss is indeed observed in numerical tests. The accuracy result also applies to spectral element method on any curvilinear mesh that can be smoothly mapped to a rectangular mesh, e.g., solving a wave equation on an annulus region with a curvilinear mesh generated by polar coordinates.
2) Monotonicity: consider solving the Poisson equation, then a scheme is called monotone if the inverse of the stiffness matrix is entrywise non-negative. It is well known that second order centered difference or P1 finite element method can form an M-matrix thus they are monotone, and high order accurate schemes in general are not monotone. But on structured meshes, high order accurate schemes can be monotone, though they do not form M-matrices. In particular, we have proven that the fourth order accurate FD scheme (Q^2 spectral element method) is a product of two M-matrices thus monotone for a variable coefficient diffusion operator: this is the first time that a high order accurate scheme is proven monotone for a variable coefficient operator. We have also proven the fifth order accurate FD scheme (Q^3 spectral element method) is a product of three M-matrices thus monotone for the Poisson equation: this is the first time that a fifth order accurate discrete Laplacian is proven monotone in two dimensions (all previously known high order monotone discrete Laplacian in 2D are fourth order accurate).
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
Whether or not the solution to the $2\frac{1}{2}$-dimensional Hall-magnetohydrodynamics system starting from any smooth initial data preserves its regularity for all time remains a challenging open problem. Although the research direction on component reduction of regularity criteria for Navier-Stokes equations and magnetohydrodynamics system has caught much attention recently, the Hall term has presented many difficulties. In this manuscript we discover a certain cancellation within the Hall term and obtain various new regularity criteria: first, in terms of a gradient of only the third component of the magnetic field; second, in terms of only the third component of the current density; third, in terms of only the third component of the velocity field; fourth, in terms of only the first and second components of the velocity field. As another consequence of the cancellation that we discovered, we are able to prove the global well-posedness of the $2\frac{1}{2}$-dimensional Hallmagnetohydrodynamics system with hyper-diffusion only for the magnetic field in the horizontal direction; we also obtained an analogous result in the 3-dimensional case via the discovery of additional cancellations. These results extend and improve various previous works. This is the joint work with Prof. Kazuo Yamazaki.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
Applications of atmospheric re-entry and geophysical flows are characterized by a large variety of separate models for specific tasks. This model variety poses significant difficulties both for the analysis and for the numerical solution. We thus need to rethink mathematical modelling and model order reduction for future numerical simulations.
In this talk, I will introduce hierarchical moment models as a flexible way to derive hierarchies of models in fluid dynamics and other applications. The general derivation procedure of the reduced models results in structural similarities of the models, which facilitate physical insight, model adaptivity, and the development of suitable numerical methods. Based on kinetic equations and shallow flows, I will exemplify the hierarchical moment approach and highlight runtime and accuracy improvements.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
Many scientific and engineering applications require solutions of partial differential equations for a wide range of parameter values (e.g. in statistical inverse problem, optimal control etc.). Additionally, many applications require inexpensive solution of PDEs, especially in a real-time context. Traditional numerical methods for the solution of PDEs (e.g. finite difference, finite element, etc.) involve a large number of unknowns and hence are unsuitabe for such applications. Reduced basis methods are a form of model order reduction that offers the potential to decrease the dimension of the problem and hence solutions are constructed with low computational cost. However, in most reduced basis methods the accuracy of a reduced basis solution is typically measured in reference to a full-order finite element solution. Often the accuracy of the full order finite element solution is itself heavily dependent on the value of parameters for certain problems, resulting in an error estimate for the reduced basis solution that is often overly optimistic.
In the first part of this talk, we present a reduced basis method with a sharp error estimate with respect to the exact solution of the PDE. A crucial element in developing such a method is the least-squares finite element method (LSFEM). LSFEMs are widely used for the solution of PDEs arising in many applications in science and engineering. LSFEMs are
based on minimizing the residual of the PDE in an appropriate norm, and have a number of attractive properties. In particular, the property relevant to this work is that these methods provide a robust and inexpensive a posteriori
error estimate with respect to the true solution. This estimate is utilized in developing the Least-Squares Reduced Basis Method presented in this talk.
The second part of the talk concerns a key ingredient in error estimates for variational problems and reduced basis methods: the so-called coercivity or inf-sup constant of the continuous problem. We characterize the coercivity constant as a spectral value of a self-adjoint linear operator; for several differential equations, we show that the coercivity constant is related to the eigenvalue of a compact operator. For these applications, convergence rates are derived and verified with numerical examples.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
Calcium dynamics in neurons containing an endoplasmic reticulum are governed by diffusion-reaction equations on interface-separated domains. Interface conditions are typically described by systems of ordinary differential equations that provide fluxes across the interfaces. For the model with ODE-flux boundary condition, we prove the existence, uniqueness, and boundedness of the solution. Then we propose and analyze an efficient implicit-explicit finite element scheme which is implicit for the parabolic operator and explicit for the nonlinear terms. We show that the stability does not depend on the spatial mesh size. Also, the optimal convergence rate in H^1 norm is obtained. We further develop a stable high-order multi-step scheme to overcome the instability and low accuracy of the previous method. Parallel algorithms are implemented for coupled PDEs on interface-separated domains. The newly designed scheme is used to solve large-scale 3D calcium models on neurons. To our knowledge, we are the first to obtain the 3D full-cell simulations with endoplasmic reticulum inside.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
It is widely known that neural networks (NNs) are universal approximators of continuous functions. However, a less known but powerful result is that a NN can accurately approximate any nonlinear continuous operator. This universal approximation theorem of operators is suggestive of the structure and potential of deep neural networks (DNNs) in learning continuous operators or complex systems from streams of scattered data. In this talk, I will present the deep operator network (DeepONet) to learn various explicit operators, such as integrals and fractional Laplacians, as well as implicit operators that represent deterministic and stochastic differential equations. I will also present several extensions of DeepONet, such as DeepM&Mnet for multiphysics problems, DeepONet with proper orthogonal decomposition (POD-DeepONet), MIONet for multiple-input operators, and multifidelity DeepONet. More generally, DeepONet can learn multiscale operators spanning across many scales and trained by diverse sources of data simultaneously. I will demonstrate the effectiveness of DeepONet and its extensions to diverse multiphysics and multiscale problems, such as nanoscale heat transport, bubble growth dynamics, high-speed boundary layers, electroconvection, and hypersonics.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied
Recent experimental studies in atomically thin materials such as graphene have offered insights into the collective motion of electrons in 2D. A wealth of intriguing optical phenomena can arise in these systems because of the coupling of the electron motion with incident electromagnetic fields.
In many applications of photonics at the nanoscale, 2D materials such as graphene may behave as conductors, and allow for the excitation and propagation of electromagnetic waves with surprisingly small length scales. These surface waves are tightly confined to the material. They can possibly beat the optical diffraction limit, in the sense that the wavelength of the excited surface waves can be much smaller than that of the incident wave in a frequency range of practical interest. A broad goal in mathematical modeling is to understand how distinct kinetic regimes of 2D electron transport can be probed, and even controlled, by electromagnetic signals.
In this talk, I will discuss recent work in describing the dispersion of electromagnetic modes that may propagate along edges of flat, anisotropic conducting sheets. Some emphasis will be placed on the role of the fractional Laplacian in this context. An emergent concept of topological character pertaining to the existence of such modes will be presented. The starting point is a boundary value problem for Maxwell’s equations coupled with the physics of the moving electrons in 2D structures. This problem will be converted to singular integral equations of the Wiener-Hopf type, and be solved explicitly. Some extensions of the analysis will be discussed, if time permits.
Please attend this week's Applied Math seminar at 4 PM Wednesday via this Zoom link.
Meeting ID: 976 3095 1027
Passcode: applied