Styxis
Company
Knowledge

Validate Dask complex Cholesky with conjugate-transpose reconstruction residuals

Check complex matrix factors against the original equation rather than copying a transpose check from a real-valued example.

Validate Dask complex Cholesky with conjugate-transpose reconstruction residuals

Receiving an array from Cholesky does not establish a correct factorization. For a complex Hermitian matrix, the conjugate transpose is part of the reconstruction relation, so tests using only real inputs can miss an important path. A small check needs a genuine imaginary component and direct inspection of upper and lower factors. [1, 2, 3]

Scope of the source report

A public Dask report describes inconsistent reconstruction residuals and upper-versus-lower results for Cholesky on a complex Hermitian positive-definite matrix. It provides an example with a fixed seed and partition sizes. This is the reporter's diagnostic material, not a reproduction or completed repair verified in the same numerical environment for this article. [1]

What to inspect in the input

Fix a small Hermitian matrix with an imaginary component and substitute both lower and upper results into their original equations. Record the reconstruction error using the conjugate transpose, normalized by the input norm. When comparing a verified complex factorization implementation, use the same equation and norm definition to avoid comparing different measures. [1, 3]

A small acceptance example

The illustrative approved budget is a normalized residual of 0.000001. An independently computed reconstruction residual of 0.00000001 passes, while 0.1 fails. These values are explanatory choices; actual tolerance depends on the matrix condition number and numerical environment. A real symmetric comparison alone cannot validate the complex path. [1, 3]

The boundary passed to Truthound

Truthound Core's BetweenValidator can check an externally computed normalized_residual between 0 and the approved budget. The operator supplies the reconstruction, norm calculation, and budget. This component evaluates a residual condition; it neither repairs Cholesky nor guarantees stability for every input. [1, 2, 3]

Application conditions

Specify conjugate-transpose reconstruction and the norm definition, then compare externally computed residuals with an approved budget.

Alternatives

Compare a verified complex factorization implementation and apply the same original-equation check to each output.

Limitations

  • The threshold is illustrative and must be chosen for the problem's condition number and numerical environment.

Evidence

Event checked
Subject checked
Modified