How reliable is a computer-based proof? (Q911568): Difference between revisions

As more and more mathematicians are realizing the power of a computer, maybe we should start considering its limitations. Even if the computer played only a minor role in a proof, the result may not be absolute. Computing errors can be broadly classified into two categories: human errors and hardware errors. Human errors are the most common; they are usually reproducible, and they are almost unavoidable. Hardware errors are random and rarely unnoticed. Human errors are the major source of our headache, but they are also under our control. Eliminating all human errors may be too ambitious. This article outline four methods to increase the confidence of computer-based results, namely: 1. prove the results by hand, 2. double checking the result, 3. consistency checking, and 4. use well-tested programs. A paper containing a computer-based proof creates a new challenge for our refereeing system. It is very difficult for a reference to check its correctness. This article proposes to treat computer-based proofs as experimental results. It also emphasizes the importance of independent verification. Furthermore, it suggests that a journal that publishes a computer-based proof has a more obligation to publish an independent verification of the result. It concludes by stating: ``As physicists learned to live with uncertainty, so we should learn to live with an `uncertain' proof''.