Not to mention, there isn't sufficient time to check a 40+ page proof.

Conference submissions using LLMs can look plausibly correct on a first pass but have a bug hiding deep in the submission. Generally, I don't think of the review process as adversarial, because the author has probably thought about the proof very carefully. Should we change how we review papers?

logical error rate of the neural decoder as a function of physical error rate. The power law fit is a smaller exponent at lower physical error rates than at higher physical error rates.

This is a good example of why you should be cautious when extrapolating to the low logical error rate regime
arxiv.org/abs/2604.08358

How large is too large for a figure to be included in a paper?

I've never thought about the viability of fitting my carry on luggage into a CRJ overhead bin, but I will from now on

GitHub - Timeroot/Lean-QuantumInfo: Quantum information theory in Lean 4 Quantum information theory in Lean 4. Contribute to Timeroot/Lean-QuantumInfo development by creating an account on GitHub.

Alexander Meiburg has been working on formal verification of results in quantum information. It's exciting to see a growing library of tools for formal proofs in QI github.com/Timeroot/Lea...

Though, it's true that we should be much more mindful of the rate of the outer code in a concatenated scheme

LDPC does reduce the effects of hook errors quite a bit. Noise-biasing the syndrome extraction circuit does help, but it might not be sufficient if a rare large error event (e.g. ionizing radiation) blows up a surface code patch mid-cycle. That said, we don't have any good models of this.

Quantum LDPC Codes of Almost Linear Distance via Homological Products We present new constructions of quantum codes of linear or close-to-linear distance and dimension with low-weight stabilizers. Only a few constructions of such codes were previously known, and were pr...

To my understanding, the techniques in this paper (Golowich and Guruswami) use products of higher chain complexes although they require qLTCs arxiv.org/abs/2411.03646

I also want to advertise a recent result by Bergamaschi and Gidney that shows constant space overhead quantum computation can be accomplished with the connectivity of a line (2502.16132). 9/9

These sorts of constructions also have good resilience against rare error bursts as long as they are sufficiently spatially localized, so we may see additional reasons to use them as we understand our error sources better. 8/9

The Yoked Surface codes by Gidney, Newman, Brooks, and Jones (2312.04522) is a related non-asymptotic approach that shows this style of memory has good practical potential. 7/9

We also are interested in making our construction practically relevant and provide a recipe for constructing hierarchical codes / syndrome extraction circuits given an input qLDPC code. This recipe can take advantage of any available long-range connectivity, but it doesn't require it. 6/9

We also saturate (up to log factors) a tradeoff between the volume of syndrome extraction circuits and the desired rate of logical error suppression by Baspin, Fawzi, and Shayeghi (2302.04317). 5/9

To our knowledge, this is the first proof that we can outperform the surface code in the asymptotic regime using purely geometrically local circuits. 4/9

We show that, at a small (log) cost, a quantum memory based on good qLDPC code families can be realized with a threshold using a 2D geometrically local circuit. At the end of the day, we have a memory with rate 1/polylog(N) and error suppression exp(-N/polylog N), exceeding the BPT bound. 3/9

Given known no-go bounds such as the Bravyi-Pouli-Terhal bound, a natural question is whether we can realize the benefits of modern qLDPC code constructions subject to locality constraints. 2/9

Happy to share that some previous work with @krishnanirudh.bsky.social and @preskill.bsky.social that we call hierarchical codes (2303.04798) was recently published in Quantum! 1/9

@quantumearl.bsky.social seeking bids for QEC26! QEC is exploding, so there’s a good case to move to an annual conference format

Has anyone thought about non-asymptotic bounds on code parameters that include the check weight? Especially near-term it would be useful to know if there's something better out there (or what weight-5, weight-6 checks buys you). I'm happy if it's "solve this crazy LP/SDP."

I suddenly regret not doing this

Oftentimes, one might write a paper that relies on a bunch (~5-10 pages) of definitions from a different paper. At what point do you say the equivalent of "import otherPaper.definitions"? If it's your own paper, then presumably the original definitions were best?

Reliable information storage in memories designed from unreliable components This is the first of two papers which consider the theoretical capabilities of computing systems designed from unreliable components. This paper discusses the capabilities of memories; the second pape...

I wonder if the early pioneers of classical fault-tolerance ever imagined where the field might go ieeexplore.ieee.org/document/677...

Happy QIP tutorials day

and I think we can all agree, 0.5" margins are probably not reasonable.

Letting near-failures through risks setting a precedent that people will just ignore the minimum requirements

Perhaps we should specify minimum margins and font size in conference submissions and automatically reject non-conformant submissions?

Can I make the obnoxious suggestion that nix might be the answer?

I think it would be really cool if someone could write a proof of a fault-tolerant threshold theorem in Lean, but it seems like a major uphill battle since it would be from scratch

I think I've also ended up with a sort of tensor network-y conclusion with decorated edges / vertices and labeled "legs." I have a description as a labeled graph where the labels satisfy some consistency conditions, but this is a bit unsatisfying since inconsistent data is possible to describe