Cholesky factorizations over the hemireals can be computed for arbitrary symmetric matrices, including singular and indefinite matrices. For singular matrices, the behavior is reminiscent of the singular value decomposition, but the performance is much better.

After creating a symmetric matrix A, compute its Cholesky factorization over the hemireal numbers like this:

F = cholfact(PureHemi, A)

Then you can use F to solve equations, e.g.,

x = F\b

If A is singular, this should be the least-squares solution.

You can compute F*F' or say rank(F). You can also convert F into matrix form with convert(Matrix, F).

To support all operations, you need to be running at least a version of julia-0.5-dev that is current with master as of 2015-12-11. However, many operations also work on julia-0.4.

F = cholfact(PureHemi, A, δ; blocksize=default)

where:

• δ is the tolerance on the diagonal values of A during factorization; any with magnitudes smaller than δ will be treated as if they are 0.
• blocksize controls the performance of the factorization algorithm.