ldl#

Purpose#

Returns the L and D factors of the LDL’ (or LDLT) factorization of a real symmetric matrix.

Format#

{ L, D } = ldl(A)#

Parameters:

A (NxN matrix) – Real symmetric data matrix

Returns:

L (NxN matrix) – Permuted lower triangular matrix, containing the factor L.
D (NxN matrix) – Block diagonal matrix containing the factor D.

Examples#

Basic Usage#

// Assign A matrix
A = { 5   9   3   4,
      9  -6   8   1,
      3   8   2   3,
      4   1   3   9 };

// Factorize matrix 'A'
{ L, D } = ldl(A);

A_new = L * D *  L';

After the code above:

       -1.50     1.00     0.00     0.00
L =     1.00     0.00     0.00     0.00
       -1.33     0.81     1.00     0.00
       -0.17     0.30    -0.25     1.00

       -6.00     0.00     0.00     0.00
D =     0.00    18.50     0.00     0.00
        0.00     0.00     0.50     0.00
        0.00     0.00     0.00     7.50

        5.00     9.00     3.00     4.00
A_new = 9.00    -6.00     8.00     1.00
        3.00     8.00     2.00     3.00
        4.00     1.00     3.00     9.00

Permuted L matrix#

// Create 5x5 matrix
A = { 8.2990  -2.7560   2.3840   3.4980   0.7520,
     -2.7560   2.2370  -2.7400   1.2930  -1.2740,
      2.3840  -2.7400   6.7890  -0.9610   0.1600,
      3.4980   1.2930  -0.9610   9.3570  -2.3780,
      0.7520  -1.2740   0.1600  -2.3780   2.2210 };

// Perform LDL decomposition
{ L, D } = ldl(A);

After the code above, the permuted L and diagonal D equal:

           1        0        0        0        0
     -0.3321   0.3114  -0.2380        1        0
L =   0.2873  -0.2494        1        0        0
      0.4215        1        0        0        0
      0.0906  -0.3419  -0.1297  -1.4970        1

      8.2990        0        0        0        0
      0        7.8826        0        0        0
D =   0             0   5.6139        0        0
      0             0        0   0.2394        0
      0             0        0        0   0.6006

Remarks#

Matrix factorization is the most computationally intense part of solving a system of linear equations. The factorization can be saved and reused multiple times to prevent the need to repeat the matrix factorization step. If you only need the LDLT factorization for this purpose, the combination of ldlp() and ldlsol() may be a better choice.
The LDL matrix factorization without permutation is not numerically stable for positive indefinite matrices. Therefore, this function uses the permutation strategy from Bunch and Kaufman. The permutations may result in an L matrix with elements above the diagonal.