ndarray_linalg

Module solveh

source
Expand description

Solve Hermitian (or real symmetric) linear problems and invert Hermitian (or real symmetric) matrices

Note that only the upper triangular portion of the matrix is used.

§Examples

Solve A * x = b, where A is a Hermitian (or real symmetric) matrix:

use ndarray::prelude::*;
use ndarray_linalg::SolveH;

let a: Array2<f64> = array![
    [3., 2., -1.],
    [2., -2., 4.],
    [-1., 4., 5.]
];
let b: Array1<f64> = array![11., -12., 1.];
let x = a.solveh_into(b).unwrap();
assert!(x.abs_diff_eq(&array![1., 3., -2.], 1e-9));

If you are solving multiple systems of linear equations with the same Hermitian or real symmetric coefficient matrix A, it’s faster to compute the factorization once at the beginning than solving directly using A:

use ndarray::prelude::*;
use ndarray_linalg::*;

/// Use fixed algorithm and seed of PRNG for reproducible test
let mut rng = rand_pcg::Mcg128Xsl64::new(0xcafef00dd15ea5e5);

let a: Array2<f64> = random_using((3, 3), &mut rng);
let f = a.factorizeh_into().unwrap(); // Factorize A (A is consumed)
for _ in 0..10 {
    let b: Array1<f64> = random_using(3, &mut rng);
    let x = f.solveh_into(b).unwrap(); // Solve A * x = b using the factorization
}

Structs§

  • Represents the Bunch–Kaufman factorization of a Hermitian (or real symmetric) matrix as A = P * U * D * U^H * P^T.

Enums§

  • Upper/Lower specification for seveal usages

Traits§

  • An interface for calculating determinants of Hermitian (or real symmetric) matrix refs.
  • An interface for calculating determinants of Hermitian (or real symmetric) matrices.
  • An interface for computing the Bunch–Kaufman factorization of Hermitian (or real symmetric) matrix refs.
  • An interface for computing the Bunch–Kaufman factorization of Hermitian (or real symmetric) matrices.
  • An interface for inverting Hermitian (or real symmetric) matrix refs.
  • An interface for inverting Hermitian (or real symmetric) matrices.
  • An interface for solving systems of Hermitian (or real symmetric) linear equations.

Type Aliases§