And given that this experiment was not particularly optimized for loss, it's likely that there's a lot of room for improvement available

Thank you for clarifying, I think this is what I was trying to express as well. When one first hears 10 billion x it sounds scary, but because losses multiply, it actually only corresponds to on log-average 100x per component, which sounds much more feasible, probably similar to 95% -> 99.9% 2Q gate

Is it possible that because there are multiple components and losses multiply, each component needs to have its loss reduced by 100x, but in total it amounts to 100 dB (10 billion x)?

I didn't realize the 100x was on a decibel scale, and not just 100x!! Thank you for clarifying that 👍

From a philosophical perspective, if quantum computing were somehow actually impossible, I think magic state distillation is a likely candidate where issues would appear first. So seeing it work experimentally at all is awesome.

...also I like that one of their circuits is sourced from a tweet.

Also, a pleasant surprise to see simultaneous work by
Google quantum also showing impressive progress with the color code and dynamic surface code. Exciting times!

Part of this exciting work was done during Sunny's internship with us; if you'd like to work on similar frontier questions with us, check out our internship posting here: job-boards.greenhouse.io/queracomputi... and our full time QEC posting here: www.quera.com/careers.

This was an amazing effort from the whole team at QuEra, as well as our collaborators at Harvard and MIT. Special shout out to Pedro, John, Niki and Sergio on the experimental side, and Sunny, Casey, Chen, Kai on the theory side.

The landscape of methods to prepare magic states is also rapidly evolving, and we look forward to further exploring the best ways to generate quantum magic on our neutral atom quantum computers.

While we demonstrate improvements in logical fidelity from distillation, gate fidelities must be further improved to lower the distillation overhead and enable multiple rounds of distillation. There is still a long road ahead, but we are optimistic about the future.

Our experiment demonstrates a key building block of large-scale, universal fault-tolerant quantum computers, and we are excited to employ high-fidelity magic states in future experiments with logical algorithms.

Our experiment leverages key aspects of the neutral atom platform, such as its dynamic reconfigurability and high degree of parallel control. For example, ten d=3, or five d=5 color codes are encoded in parallel, and transversal gates also have high parallelism.

Moreover, we experimentally probe key aspects of MSD, such as its quadratic error suppression, by varying the quality of the input magic state and observing the output fidelity.

Here, we realize magic state distillation at the *logical level* with a neutral atom quantum computer. We show that the output logical magic state has a fidelity higher than the input logical magic state, for both distance 3 and 5 color codes.

Impressive experiments have demonstrated MSD with physical qubits and error-suppressed encodings of magic states, but MSD has not been demonstrated with logical qubits. The use of logical qubits is crucial, to benefit from protection against errors in distillation operations.

One of the well-established methods to prepare high-fidelity magic resource states is magic state distillation (MSD). Amazingly, one can "distill" a better magic state from multiple noisier inputs.

Unfortunately, high-quality magic states are one of the most complex things to prepare for large-scale quantum computers.

This is where magic states come in. "Magic", which describes how far away a quantum state is from a stabilizer state, is a key resource for performing universal quantum computation and achieving quantum advantage.

Stabilizer states and Clifford operations are often easy to implement on an error-corrected quantum computer. However, such states can also be efficiently simulated classically, and do not suffice for universal quantum computation.