-
Spectral/hp Hull: A Degree of Freedom Reducing Discontinuous Spectral Element Method for Conservation Laws with Application to Compressible Fluid Flow
Authors:
A. Ghasemi,
L. K. Taylor,
J. C. Newman III
Abstract:
In conventional spectral/finite element methods, the triangulation/quadrilateralization of the domain produces many interior edges which require additional DOF. What if we could directly use the original hull without going to triangulation/quadrilateralization? There are three major difficulties with this type of approach that are addressed: 1- How can a convex hull tessellation be obtained for co…
▽ More
In conventional spectral/finite element methods, the triangulation/quadrilateralization of the domain produces many interior edges which require additional DOF. What if we could directly use the original hull without going to triangulation/quadrilateralization? There are three major difficulties with this type of approach that are addressed: 1- How can a convex hull tessellation be obtained for complex geometries encountered in practical engineering applications? 2- How can basis functions and quadrature points be defined on these hulls? 3- How can this type of grid and corresponding basis functions be used in practice to yield accurate discretization of nonlinear conservation laws? Geometrical approaches to tackle the first challenge are discussed . To solve the second challenge, for the first time in the literature, a closed form relation is proposed to approximate Fekete points (a Nondeterministic Polynomial (NP) problem) on a general convex/concave polyhedral. The approximate points are used to generate basis functions using the SVD of the Vandermonde matrix. The type of basis functions derived include Lagrange basis, orthogonal and orthonormal hull basis and radial basis. It is shown that the hull basis is the best choice to enforce minimum DOF while maintaining a small Lebesgue constant when very high p-refinement is done. The proposed hull basis is rigorously proven to achieve arbitrary order of accuracy by satisfying Weierstrass approximation theorem in $\mathbb{R}^d$. The third challenge is met using a standard Discontinuous Galerkin (DG) and Discontinuous Least-Squares (DLS) spectral hull formulations. The accuracy and stability of the formulation is demonstrated for the linearized acoustics and two-dimensional compressible Euler equations on some benchmark problems including a cylinder, airfoil, vortex convection and compressible vortex shedding from a triangle.
△ Less
Submitted 24 May, 2016;
originally announced May 2016.
-
Numerical Stability and Catalan Numbers
Authors:
Arash Ghasemi,
Kidambi Sreenivas,
Lafayette K. Taylor
Abstract:
To predict allowable time-step size for the fully discretized nonlinear differential equations, a stability theory is developed using exact determination of an infinite perturbation series. Mathematical induction is used to determine the coefficients of the series. It is discovered that the closed-form equation for the nonlinear shift of generic polynomial non-linearity can be written as a series…
▽ More
To predict allowable time-step size for the fully discretized nonlinear differential equations, a stability theory is developed using exact determination of an infinite perturbation series. Mathematical induction is used to determine the coefficients of the series. It is discovered that the closed-form equation for the nonlinear shift of generic polynomial non-linearity can be written as a series expansion where the coefficients are the Pfaff-Fuss-Catalan numbers in Combinatorics. This reveals criteria which can be used to analytically determine the allowable time step. It is shown that stability region decreases when the nonlinearity of the differential equation increases. Therefore, the maximum allowable time step is severely limited by the nonlinearity even if an unconditionally stable scheme (in a linear sense) is used. The theory is applied to general system of time-dependent nonlinear Partial Differential Equations.
△ Less
Submitted 2 November, 2013; v1 submitted 18 September, 2013;
originally announced September 2013.
-
A Progressive Statistical Method for Preconditioning Matrix-Free Solution of High-Order Discretization of Linear Time-Dependent Problems
Authors:
A. Ghasemi,
L. K. Taylor
Abstract:
Preconditioning of a linear system obtained from spectral discretization of time-dependent PDEs often results in a full matrix which is expensive to compute and store specially when the problem size increases. A matrix-free implementation is usually applied to resolve this issue. In this framework, preconditioning is typically challenging since the entries of the matrix are not explicitly availabl…
▽ More
Preconditioning of a linear system obtained from spectral discretization of time-dependent PDEs often results in a full matrix which is expensive to compute and store specially when the problem size increases. A matrix-free implementation is usually applied to resolve this issue. In this framework, preconditioning is typically challenging since the entries of the matrix are not explicitly available. In this short note, we propose a statistical approach to gradually create a preconditioner matrix by collecting the information obtained from matrix-vector product in the Arnoldi loop of an unpreconditioned Krylov subspace algorithm. The gathered information are then correlated using a multiple regressors estimate where the error is assumed to be normally distributed. This procedure yields a banded diagonal matrix which is then used as a preconditioner in the next iterative solve. This is repeated between iterative solves until a good preconditioner is constructed on fly. This statistically iterative procedure is progressive since the fidelity of the preconditioning matrix improves by adding more data obtained from matrix-vector product during the entire solution procedure. The proposed algorithm is validated for a sample implementation.
△ Less
Submitted 26 August, 2013;
originally announced August 2013.
