Why use iterative methods instead of just diagonalizing the whole matrix?

Physical Hamiltonians can have dimensions in the billions but are sparse, so storing or factoring them fully is impossible. Iterative Krylov methods like Lanczos only need the action of the matrix on a vector and can extract the few lowest eigenstates that physics usually cares about.

Why does Hermiticity of physical matrices matter numerically?

Hermitian matrices have real eigenvalues and orthogonal eigenvectors, which lets specialized, more stable and efficient algorithms be used and guarantees that computed energies are real, matching the physics.

Numerical Linear Algebra and Eigenproblems in Physics

Discretizing a physical operator turns physics into matrices, and finding the energies and modes of a system becomes the numerical problem of solving large linear systems and computing eigenvalues and eigenvectors.

یافتن موضوع با PaperMindبه‌زودیFind papers & topics

Tools & resources

دریافت اسلایدها

Learn & explore

ویدیوبه‌زودی

Definition

Numerical linear algebra in physics is the set of algorithms for solving matrix equations and eigenvalue problems that arise when continuous physical operators are represented in a finite basis or on a grid.

Scope

This topic covers the matrix computations central to physics: solving linear systems by direct and iterative methods, and computing eigenvalues and eigenvectors of large, often sparse, Hermitian matrices via QR, Jacobi, Lanczos and conjugate-gradient algorithms. It emphasizes the structure of physical matrices such as sparsity and Hermiticity.

Core questions

How are large linear systems from discretized physics solved without forming dense inverses?
How are eigenvalues and eigenvectors of a Hamiltonian matrix computed numerically?
Why are iterative Krylov methods preferred for large sparse matrices over direct factorization?
How does the Lanczos algorithm extract a few extreme eigenvalues of a huge sparse Hermitian matrix?

Key theories

Direct and iterative linear solvers: Linear systems are solved either by direct factorization such as LU and Cholesky, exact up to round-off, or by iterative Krylov methods such as conjugate gradients that exploit sparsity and converge to a tolerance.
Eigenvalue algorithms: Eigenvalues and eigenvectors are computed by the QR algorithm and Jacobi rotations for dense matrices, giving the discrete spectrum of a physical operator represented in a finite basis.
Lanczos and Krylov subspace methods: The Lanczos algorithm builds a small tridiagonal projection of a large sparse Hermitian matrix in a Krylov subspace, allowing a few extreme eigenvalues and eigenvectors to be found without ever storing the full matrix.

Clinical relevance

These algorithms compute energy levels and wavefunctions in quantum mechanics, normal modes of vibration, band structures in solids, and the linear systems behind discretized field equations, making them indispensable across electronic structure and condensed-matter simulation.

History

Practical matrix eigenvalue computation matured in the mid-twentieth century with Lanczos's 1950 iteration and the QR algorithm of the early 1960s; the rise of large sparse problems in physics made Krylov subspace methods the dominant tools for spectra of high-dimensional Hamiltonians.

Key figures

Cornelius Lanczos
Gene H. Golub
James H. Wilkinson

Seminal works

golub2013
lanczos1950

Frequently asked questions

Why use iterative methods instead of just diagonalizing the whole matrix?: Physical Hamiltonians can have dimensions in the billions but are sparse, so storing or factoring them fully is impossible. Iterative Krylov methods like Lanczos only need the action of the matrix on a vector and can extract the few lowest eigenstates that physics usually cares about.
Why does Hermiticity of physical matrices matter numerically?: Hermitian matrices have real eigenvalues and orthogonal eigenvectors, which lets specialized, more stable and efficient algorithms be used and guarantees that computed energies are real, matching the physics.