I understand the definition of accuracy for a backward stable numerical algorithm, which shows how such a quantity depends both on machine-epsilon and on the condition number of the problem under examination.

However, many numerical algorithms used in practice are only stable, not backward stable. Which is a proper definition of accuracy for this class of algorithms? I feel it should still involve the notions of condition number and machine-epsilon, but I'm looking for formal definition / derivation



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.