by An Uncommon Lab

lcf2mat

Converts lower Cholesky factors into corresponding full matrices. This applies across dimensions 3 and 4 and so is appropriate to use on sets of Cholesky factors, such as those returned by Monte-Carlo tests, where the third dimension represents the sample index and the fourth is the run index.

Inputs

Outputs

Example: Basic Case

Let's create an initial covariance matrix.

P = randcov(3)

P =
    0.5052   -0.2311   -0.2059
   -0.2311    0.5124   -0.0714
   -0.2059   -0.0714    0.7502

Now we calculate the lower Cholesky factor.

C = sqrtpsdm(P, 'lower');

We can manually construct the covariance or use lcf2mat.

C * C.'
lcf2mat(C)

ans =
    0.5052   -0.2311   -0.2059
   -0.2311    0.5124   -0.0714
   -0.2059   -0.0714    0.7502
ans =
    0.5052   -0.2311   -0.2059
   -0.2311    0.5124   -0.0714
   -0.2059   -0.0714    0.7502

Example: Returning Only the Top Left Submatrix

Sometimes, for efficiency, we need to return only the top n-by-n submatrix of matrix that is reconstructed from the lower Cholesky factor. We can do this with the second argument.

P = randcov(5);
C = sqrtpsdm(P, 'lower');
P(1:3, 1:3)
lcf2mat(C, 3)

ans =
    0.6846   -0.1233   -0.0311
   -0.1233    0.4337    0.1801
   -0.0311    0.1801    0.5882
ans =
    0.6846   -0.1233   -0.0311
   -0.1233    0.4337    0.1801
   -0.0311    0.1801    0.5882

Example: For Multiple Samples and Runs

Let's create covariance matrices for each of 10 samples on each of 5 runs to mimic results we might obtain from a Monte-Carlo test.

P = zeros(3, 3, 10, 5);
C = zeros(3, 3, 10, 5);
for s = 1:10
    for r = 1:5
        P(:, :, s, r) = randcov(3);
        C(:, :, s, r) = sqrtpsdm(P(:, :, s, r), 'lower');
    end
end

Let's show that we can reconstruct the covariance matrices from the various lower Cholesky factors all at once and compare those results to the original covariance matrices.

P2 = lcf2mat(C);
max(abs(P(:) - P2(:)))

ans =
   1.1102e-16

See Also

sqrtpsdm, ldfactor.m, udfactor.m