← All notes & connection map

SEM

Spectral Element Method

Created on: 2025-01-17
Tags: #SEM
Course: NEK5000


FEM

  • Finite Element Method involves a mesh with range of cell sizes dependent on problems.
  • It is computationally inexpensive as compared to FDM due to the use of Cartesian Mesh.
  • Although, meshing is a science in itself.

1D Wave Equation

\(\rho(x) \ddot{u}(x,t) - \partial_{x}[\mu(x)\partial_{x}u(x,t)] = f(x,t)\)

  1. $\partial_{x}u(x,t) = 0$ at $x=0,L$
  2. $u(x,0)=\dot{u}(x,0)=0$

Compact Subset: A generalization of a space being closed.

Weak Form of the 1D Wave Equation

\(\int_{G} w \rho \ddot{u}\,dx - \int_{G} w \partial_{x}[\mu\partial_{x}u]\,dx = \int_{G}w f\,dx\) Where $w$ is the test function.

Applying integration by part to the above equation, we get \(\int_{G} w \rho \ddot{u}\,dx + \int_{G}\mu\partial_{x}w\,\partial_{x}u\,dx = \int_{G}w f\,dx\) The essential boundary condition is already satisfied by the weak form.

Galerkin Projection

We will approximate $u(x,t)$ by a finite superposition of $n$ basis functions $\psi_{i}$. \(\bar{u}(x,t) = \sum_{i=1}^n U_{i} \psi_{i}(x) \approx u(x,t)\)

Finite Element Form

Replacing the approximate value in the weak form

\[\sum_{i=1}^n \left[ \ddot{U}_{i}(t)\int_{G}\rho \psi_{i} \psi_{j} dx \right] + \sum_{i=1}^n \left[ U_{i}(t)\int_{G}\mu\, \partial_{x}\psi_{j} \partial_{x}\psi_{i} dx \right] = \int \psi_{j} f\, dx\] \[M\,\ddot{\vec{U}}(t) + K\,\vec{U}(t) = \vec{f}(t)\]

Where,

  1. M is the mass matrix, given by \(M_{ij} = \int_{G}\rho \psi_{i} \psi_{j} dx\)
  2. K is the stiffness matrix, given by \(K_{ij} = \int_{G}\mu\, \partial_{x}\psi_{j} \partial_{x}\psi_{i} dx\)
  3. f is the source term, given by \(f_{j} = \int \psi_{j} f\, dx\)

Elements

  • In 2D, the domains $G_{i}$ should be disjoint such that \(\bigcup_{i=1}^n G_{i} = G\) \(G_{i} \cap G_{j} = \varnothing, \,\,\, \forall\, i\ne j\)
  • For each element, we define \(M^e \ddot{U}^e(t) + K^e U^e(t) = f^e(t)\)

Basis Functions

  • We choose $n+1$ Lagrange polynomials with Gauss-Lobatto-Legendre (GLL) points as collocation points.
  • $u_{i}$ are the nodal values at specific GLL points.
  • GLL points are inside the interval except 2 points are on the ends to have a better curvature.
  • For a $N^{th}$ polynomial basis, we use $N+1$ basis functions.
  • Integration using Gauss Quadrature. \(\psi_{i}[x(\zeta)] = L^N_{i}(\zeta), \,\,\, \zeta=[-1,1]\)

Key Differences Between SEM and FEM

  1. Polynomial Order:
    • FEM: Uses low-order polynomial basis functions (e.g., linear, quadratic).
    • SEM: Uses high-order polynomial basis functions $(N \geq 3)$, leading to spectral accuracy for smooth problems.
  2. Integration Points:
    • FEM: Typically uses uniform quadrature points (e.g., Gauss quadrature).
    • SEM: Uses Gauss-Lobatto-Legendre (GLL) points, derived from Legendre polynomials, for both basis function interpolation and numerical integration.

Linked from

  • No other published notes link here yet.