Spectral Method

Spectral methods are a class of numerical techniques that solve differential equations and optimization problems by expanding functions in orthogonal bases — Fourier series, Chebyshev polynomials, or eigenfunctions of an appropriate operator. Unlike finite-difference or finite-element methods, which approximate derivatives locally, spectral methods exploit global smoothness: a function that is smooth everywhere can be represented with exponential accuracy by a small number of basis coefficients.

The connection to spectral graph theory is more than terminological. Both fields leverage the eigenstructure of linear operators — the Laplacian in graph theory, the differential operator in numerical analysis — to transform complex problems into simpler ones. The spectral gap determines the conditioning of spectral methods: a small gap means slow convergence and numerical instability; a large gap permits rapid, accurate solution.

Spectral methods achieve their power by changing the representation of the problem, not by refining the discretization. They are the computational embodiment of a deep principle: the right basis makes the hard problem easy. In harmonic analysis, in spectral clustering, and in quantum chemistry, the same insight recurs — find the natural modes of the system, and compute in their coordinates.