Error suppression is like learning the keyboard on your laptop so that you press the keys you want to press. Error correction is like seeing a mistake you made anyway, tapping backspace a few times to undo it, and then typing the correct characters. But despite your best efforts, you just can’t avoid making mistakes entirely. After all, the CAPTCHA says you’re only human. Error mitigation is like using Grammarly to catch the mistakes that slipped past you.
With quantum computing, we apply error suppression techniques to try to prevent errors from occurring. They happen anyway, so we’re developing error correction techniques to detect and correct those. But we’re not there yet, so we’re getting imperfect results. Mitigation happens after the computation and there’s more than one technique, but it’s about patterns. A pattern of errors, called a noise model, can be used to apply corrective adjustments to the final computation, hopefully providing an accurate result.
Error mitigation is useful on a small scale, but it has limitations. It lengthens a computation because the error patterns are continuously changing and it takes time to build a current noise model. Longer circuits take even more time because there are more places for the errors to hide; think of it as the difference between playing hide-and-seek in a cottage versus a mansion. And these techniques stop working as the count of qubits grows beyond a rather small number, so they’ll only be useful during the NISQ era.