René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"better symmetry check test","articleBody":"Hi,\n\nWhile porting our MATLAB code to Python using the control library, we noticed that the current symmetry check:\n\n```\ndef _is_symmetric(M):\n    M = np.atleast_2d(M)\n    if isinstance(M[0, 0], inexact):\n        eps = finfo(M.dtype).eps\n        return ((M - M.T) \u003c eps).all()\n    else:\n        return (M == M.T).all()\n```\n\nhas some limitations:\n\n1. It performs elementwise comparison without using abs(), so small negative differences (e.g., from roundoff) can cause false failures. \n2. It uses a fixed eps, which is stricter than MATLAB's behavior and doesn’t scale with matrix magnitude.\n\nWould it be possible to adopt a more numerically robust check, similar to what SciPy uses?\n\n```\nfor ind, mat in enumerate((q, r)):\n    if norm(mat - mat.conj().T, 1) \u003e np.spacing(norm(mat, 1)) * 100:\n        raise ValueError(f\"Matrix {'qr'[ind]} should be symmetric/hermitian.\")\n```\n\nThis approach allows for small, scale-aware numerical asymmetries and works well for both real symmetric and complex Hermitian matrices.\n\nPrithvi","author":{"url":"https://github.com/dakeprithvi","@type":"Person","name":"dakeprithvi"},"datePublished":"2025-07-23T18:11:07.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":10},"url":"https://github.com/1174/python-control/issues/1174"}