Merge pull request #1050 from metric-space/polar-decomposition-take-2

Take-2 of polar-decomposition
This commit is contained in:
Sébastien Crozet 2021-12-31 09:57:08 +01:00 committed by GitHub
commit 99ac8c4032
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 100 additions and 2 deletions

View File

@ -1,8 +1,8 @@
use crate::storage::Storage; use crate::storage::Storage;
use crate::{ use crate::{
Allocator, Bidiagonal, Cholesky, ColPivQR, ComplexField, DefaultAllocator, Dim, DimDiff, Allocator, Bidiagonal, Cholesky, ColPivQR, ComplexField, DefaultAllocator, Dim, DimDiff,
DimMin, DimMinimum, DimSub, FullPivLU, Hessenberg, Matrix, RealField, Schur, SymmetricEigen, DimMin, DimMinimum, DimSub, FullPivLU, Hessenberg, Matrix, OMatrix, RealField, Schur,
SymmetricTridiagonal, LU, QR, SVD, U1, UDU, SymmetricEigen, SymmetricTridiagonal, LU, QR, SVD, U1, UDU,
}; };
/// # Rectangular matrix decomposition /// # Rectangular matrix decomposition
@ -17,6 +17,7 @@ use crate::{
/// | LU with partial pivoting | `P⁻¹ * L * U` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` is a permutation matrix. | /// | LU with partial pivoting | `P⁻¹ * L * U` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` is a permutation matrix. |
/// | LU with full pivoting | `P⁻¹ * L * U * Q⁻¹` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` and `Q` are permutation matrices. | /// | LU with full pivoting | `P⁻¹ * L * U * Q⁻¹` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` and `Q` are permutation matrices. |
/// | SVD | `U * Σ * Vᵀ` | `U` and `V` are two orthogonal matrices and `Σ` is a diagonal matrix containing the singular values. | /// | SVD | `U * Σ * Vᵀ` | `U` and `V` are two orthogonal matrices and `Σ` is a diagonal matrix containing the singular values. |
/// | Polar (Left Polar) | `P' * U` | `U` is semi-unitary/unitary and `P'` is a positive semi-definite Hermitian Matrix
impl<T: ComplexField, R: Dim, C: Dim, S: Storage<T, R, C>> Matrix<T, R, C, S> { impl<T: ComplexField, R: Dim, C: Dim, S: Storage<T, R, C>> Matrix<T, R, C, S> {
/// Computes the bidiagonalization using householder reflections. /// Computes the bidiagonalization using householder reflections.
pub fn bidiagonalize(self) -> Bidiagonal<T, R, C> pub fn bidiagonalize(self) -> Bidiagonal<T, R, C>
@ -186,6 +187,62 @@ impl<T: ComplexField, R: Dim, C: Dim, S: Storage<T, R, C>> Matrix<T, R, C, S> {
{ {
SVD::try_new_unordered(self.into_owned(), compute_u, compute_v, eps, max_niter) SVD::try_new_unordered(self.into_owned(), compute_u, compute_v, eps, max_niter)
} }
/// Computes the Polar Decomposition of a `matrix` (indirectly uses SVD).
pub fn polar(self) -> (OMatrix<T, R, R>, OMatrix<T, R, C>)
where
R: DimMin<C>,
DimMinimum<R, C>: DimSub<U1>, // for Bidiagonal.
DefaultAllocator: Allocator<T, R, C>
+ Allocator<T, DimMinimum<R, C>, R>
+ Allocator<T, DimMinimum<R, C>>
+ Allocator<T, R, R>
+ Allocator<T, DimMinimum<R, C>, DimMinimum<R, C>>
+ Allocator<T, C>
+ Allocator<T, R>
+ Allocator<T, DimDiff<DimMinimum<R, C>, U1>>
+ Allocator<T, DimMinimum<R, C>, C>
+ Allocator<T, R, DimMinimum<R, C>>
+ Allocator<T, DimMinimum<R, C>>
+ Allocator<T::RealField, DimMinimum<R, C>>
+ Allocator<T::RealField, DimDiff<DimMinimum<R, C>, U1>>,
{
SVD::new_unordered(self.into_owned(), true, true)
.to_polar()
.unwrap()
}
/// Attempts to compute the Polar Decomposition of a `matrix` (indirectly uses SVD).
///
/// # Arguments
///
/// * `eps` tolerance used to determine when a value converged to 0 when computing the SVD.
/// * `max_niter` maximum total number of iterations performed by the SVD computation algorithm.
pub fn try_polar(
self,
eps: T::RealField,
max_niter: usize,
) -> Option<(OMatrix<T, R, R>, OMatrix<T, R, C>)>
where
R: DimMin<C>,
DimMinimum<R, C>: DimSub<U1>, // for Bidiagonal.
DefaultAllocator: Allocator<T, R, C>
+ Allocator<T, DimMinimum<R, C>, R>
+ Allocator<T, DimMinimum<R, C>>
+ Allocator<T, R, R>
+ Allocator<T, DimMinimum<R, C>, DimMinimum<R, C>>
+ Allocator<T, C>
+ Allocator<T, R>
+ Allocator<T, DimDiff<DimMinimum<R, C>, U1>>
+ Allocator<T, DimMinimum<R, C>, C>
+ Allocator<T, R, DimMinimum<R, C>>
+ Allocator<T, DimMinimum<R, C>>
+ Allocator<T::RealField, DimMinimum<R, C>>
+ Allocator<T::RealField, DimDiff<DimMinimum<R, C>, U1>>,
{
SVD::try_new_unordered(self.into_owned(), true, true, eps, max_niter)
.and_then(|svd| svd.to_polar())
}
} }
/// # Square matrix decomposition /// # Square matrix decomposition

View File

@ -641,6 +641,28 @@ where
} }
} }
} }
/// converts SVD results to Polar decomposition form of the original Matrix: `A = P' * U`.
///
/// The polar decomposition used here is Left Polar Decomposition (or Reverse Polar Decomposition)
/// Returns None if the singular vectors of the SVD haven't been calculated
pub fn to_polar(&self) -> Option<(OMatrix<T, R, R>, OMatrix<T, R, C>)>
where
DefaultAllocator: Allocator<T, R, C> //result
+ Allocator<T, DimMinimum<R, C>, R> // adjoint
+ Allocator<T, DimMinimum<R, C>> // mapped vals
+ Allocator<T, R, R> // result
+ Allocator<T, DimMinimum<R, C>, DimMinimum<R, C>>, // square matrix
{
match (&self.u, &self.v_t) {
(Some(u), Some(v_t)) => Some((
u * OMatrix::from_diagonal(&self.singular_values.map(|e| T::from_real(e)))
* u.adjoint(),
u * v_t,
)),
_ => None,
}
}
} }
impl<T: ComplexField, R: DimMin<C>, C: Dim> SVD<T, R, C> impl<T: ComplexField, R: DimMin<C>, C: Dim> SVD<T, R, C>

View File

@ -153,6 +153,25 @@ mod proptest_tests {
prop_assert!(relative_eq!(&m * &sol2, b2, epsilon = 1.0e-6)); prop_assert!(relative_eq!(&m * &sol2, b2, epsilon = 1.0e-6));
} }
} }
#[test]
fn svd_polar_decomposition(m in dmatrix_($scalar)) {
let svd = m.clone().svd_unordered(true, true);
let (p, u) = svd.to_polar().unwrap();
assert_relative_eq!(m, &p* &u, epsilon = 1.0e-5);
// semi-unitary check
assert!(u.is_orthogonal(1.0e-5) || u.transpose().is_orthogonal(1.0e-5));
// hermitian check
assert_relative_eq!(p, p.adjoint(), epsilon = 1.0e-5);
/*
* Same thing, but using the method instead of calling the SVD explicitly.
*/
let (p2, u2) = m.clone().polar();
assert_eq!(p, p2);
assert_eq!(u, u2);
}
} }
} }
} }