Gauss-Lobatto quadrature

             

  • 本词条需要更多讲解,便于帮助理解.
Prerequisite definite integral

   Gauss quadrature can be approximated by summation

\begin{equation} \int f(x) \,\mathrm{d}{x} \approx \sum_i w_i f(x_i) \end{equation}
For a set of $x_i, w_i$ in a certain interval, if it is of degree $N$, then when $f(x)$ is a polynomial of degree less than or equal to $N$, the above formula is exactly true.

Gauss-Lobatto points

   In Gauss-Lobatto integration, let the number of sampling points (including two end points) be $N$. If the integrand is a polynomial of order $2N-3$ ($f(x) = x^{2n-3} + \dots$), there is no error in the integration.

   Note that the Gauss-Lobatto integral is symmetric

\begin{equation} x_j = -x_{N-j+1} \qquad w_{0j} = w_{0,N-j+1} \end{equation}

Orthogonal normalized basis

   Each base is a polynomial of order $N-1$. Since the order is the same as the number of zeros, the polynomial can be uniquely determined, that is, the Lagrange interpolation polynomial

\begin{equation} \begin{aligned} p_n(x) &= \prod_{i=1}^{n-1} \frac{x-x_i}{x_n-x_i} \prod_{i=n+1}^{N} \frac{x-x_i}{x_n-x_i}\\ &= \frac{x-x_1}{x_n-x_1} \times\dots\times\frac{x-x_{n-1}}{x_n-x_{n-1}}\frac{x-x-{n+1}}{x_n-x_{n+1}} \dots \frac{x-x_N}{x_n-x_N} \end{aligned} \end{equation}
\begin{equation} p_n(x_{n'}) = \delta_{n, n'} \end{equation}

   Since any two base products are only polynomials of order $2N-2$, the sum can be used to replace their integrals.

\begin{equation} \int_{-1}^1 p_i(x) p_j(x) \,\mathrm{d}{x} = \sum_k w_k p_i(x_k) p_j(x_k) = w_i \delta_{ij} \end{equation}

   So $N$ orthogonal normalization basis is

\begin{equation} f_n(x) = \frac{1}{\sqrt{w_n}} p_n(x) \end{equation}
Satisfy
\begin{equation} f_i(x_j) = \frac{1}{\sqrt{w_i}} \delta_{ij} \end{equation}

   Another equivalent representation method is to use the polynomial corresponding to the Gauss-Lobatto numerical integration of order $N$, the derivative of Legendre polynomial of order $N-1$, $P'_{N-1}(y)$, to construct a polynomial that satisfies the conditions. According to the definition, its $N-2$ zeros are respectively $y_2, y_3\dots y_{N-1}$. In order to add the two zeros of $y_1=-1$ and $y_n=1$, they are transformed into a $N$ degree polynomial

\begin{equation} (1-y^2)P'_{N-1}(y) \end{equation}
However, eq. 2 requires $p_n(y_n)=1$, so we divide eq. 8 by its own tangent at $y_n$ and form the limit type $0/0=1$ at $y=y_n$ to get the order polynomial $u_n(y)$.
\begin{equation} u_n(y) = \frac{(1-y^2)P'_{N-1}(y)}{[(1-y^2)P'_{N-1}(y)]'_y} = y_n (y-y_n) \end{equation}
This formula is actually the same polynomial as eq. 3 , because all polynomials of order $N-1$ with $N-1$ zeros can be factored into eq. 3 multiplied by an undetermined constant. Use eq. 9 to quickly expand polynomials (because the coefficients of Legendre polynomials can be directly calculated by formulas).

         

© 小时科技 保留一切权利