nalgebra/linalg/
balancing.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
//! Functions for balancing a matrix.

use simba::scalar::RealField;
use std::ops::{DivAssign, MulAssign};

use crate::allocator::Allocator;
use crate::base::dimension::Dim;
use crate::base::{Const, DefaultAllocator, OMatrix, OVector};

/// Applies in-place a modified Parlett and Reinsch matrix balancing with 2-norm to the matrix and returns
/// the corresponding diagonal transformation.
///
/// See <https://arxiv.org/pdf/1401.5766.pdf>
pub fn balance_parlett_reinsch<T: RealField, D: Dim>(matrix: &mut OMatrix<T, D, D>) -> OVector<T, D>
where
    DefaultAllocator: Allocator<D, D> + Allocator<D>,
{
    assert!(matrix.is_square(), "Unable to balance a non-square matrix.");

    let dim = matrix.shape_generic().0;
    let radix: T = crate::convert(2.0f64);
    let mut d = OVector::from_element_generic(dim, Const::<1>, T::one());

    let mut converged = false;

    while !converged {
        converged = true;

        for i in 0..dim.value() {
            let mut n_col = matrix.column(i).norm_squared();
            let mut n_row = matrix.row(i).norm_squared();
            let mut f = T::one();

            let s = n_col.clone() + n_row.clone();
            n_col = n_col.sqrt();
            n_row = n_row.sqrt();

            if n_col.clone().is_zero() || n_row.clone().is_zero() {
                continue;
            }

            while n_col.clone() < n_row.clone() / radix.clone() {
                n_col *= radix.clone();
                n_row /= radix.clone();
                f *= radix.clone();
            }

            while n_col.clone() >= n_row.clone() * radix.clone() {
                n_col /= radix.clone();
                n_row *= radix.clone();
                f /= radix.clone();
            }

            let eps: T = crate::convert(0.95);
            #[allow(clippy::suspicious_operation_groupings)]
            if n_col.clone() * n_col + n_row.clone() * n_row < eps * s {
                converged = false;
                d[i] *= f.clone();
                matrix.column_mut(i).mul_assign(f.clone());
                matrix.row_mut(i).div_assign(f.clone());
            }
        }
    }

    d
}

/// Computes in-place `D * m * D.inverse()`, where `D` is the matrix with diagonal `d`.
pub fn unbalance<T: RealField, D: Dim>(m: &mut OMatrix<T, D, D>, d: &OVector<T, D>)
where
    DefaultAllocator: Allocator<D, D> + Allocator<D>,
{
    assert!(m.is_square(), "Unable to unbalance a non-square matrix.");
    assert_eq!(m.nrows(), d.len(), "Unbalancing: mismatched dimensions.");

    for j in 0..d.len() {
        let mut col = m.column_mut(j);
        let denom = T::one() / d[j].clone();

        for i in 0..d.len() {
            col[i] *= d[i].clone() * denom.clone();
        }
    }
}