Notes on Lagrange Interpolating Polynomials
This article offers a precise and fundamental exploration of Lagrange interpolating polynomials, a cornerstone in numerical analysis for fitting curves to data points. It meticulously details their construction, proves their existence and uniqueness, and explains their role as a basis in polynomial vector spaces. The piece stands out for its clear mathematical derivations and illustrative examples, appealing to those interested in the theoretical underpinnings of computational methods.
The Lowdown
Polynomial interpolation is a method to find a polynomial function that perfectly passes through a given set of distinct data points (nodes and values). This fundamental problem in numerical analysis can be approached in several ways, each with its own advantages and practical considerations. This article primarily focuses on the elegant method introduced by Joseph-Louis Lagrange.
- Initial Approach via Linear Algebra: The problem can be framed as a system of linear equations, where the unknown coefficients of a generic polynomial are solved using a matrix equation involving the Vandermonde matrix. While this matrix is invertible for distinct points, guaranteeing a unique solution, it is often numerically ill-conditioned in practice, making direct inversion problematic for computational accuracy.
- The Lagrange Polynomial: This method introduces 'Lagrange basis functions,' L_k(x), which are specifically constructed to be 1 at a target node x_k and 0 at all other nodes. The interpolating polynomial is then a simple linear combination of these basis functions, scaled by their corresponding y-values (P(x) = sum(y_k * L_k(x))).
- Construction of Basis Functions: Each L_k(x) is derived by multiplying (x - x_j) terms for all j != k and then normalizing the result so that L_k(x_k) = 1. The article includes visual examples of these basis functions and the resulting interpolating polynomial.
- Uniqueness and Degree: The article presents the Polynomial Interpolation Theorem, stating that for any n+1 distinct data points, there exists a unique polynomial of degree at most n that interpolates them. Uniqueness is rigorously proven by showing that if two such polynomials existed, their difference would be a zero polynomial, a contradiction for a non-zero polynomial of degree n with n+1 roots.
- Lagrange Basis for P_n: The Lagrange basis functions are shown to form an actual linear algebra basis for the vector space of all real polynomials of degree at most n (P_n). This is demonstrated by proving their linear independence and that they span the space.
- Simplified Matrix Representation: When using the Lagrange basis, the system of equations for interpolation simplifies significantly, leading to an identity matrix, which directly yields the coefficients (the y-values) without complex calculations.
- Vandermonde Matrix Appendix: An appendix provides a detailed proof of why the Vandermonde matrix is invertible when the x-coordinates are distinct, by demonstrating that its determinant is non-zero through inductive algebraic manipulation.
The Lagrange interpolating polynomial provides a clear, constructive method for polynomial interpolation, elegantly demonstrating the existence and uniqueness of such a polynomial and offering a more numerically stable alternative in many scenarios compared to direct Vandermonde matrix inversion.