Recent comments from SciRate

Indeed the fact that it uses "vertexaisearch.cloud.google.com" as a redirect to the simple arxiv link is proof that this is output from Google's Gemini AI. (Or at least, proof that someone put the link into Gemini and asked it to read it, because this is the only way to get such a redirect link.)

...(continued)

I submitted a PR fixing the issue and they merged it ( https://github.com/tequilahub/pauliengine/pull/1 ).

**ChatGPT:** this is AI-generated pseudo-audit prose attempting to pose as a new discipline (“Information Physics”). It is rhetorical cosplay, not a scientific review.

The comment is almost certainly produced by a language model wrapped in an authoritative persona. The bureaucratic scaffolding—“A

...(continued)

--- RAPPORT D'AUDIT PHYSIQUE DE L'INFORMATION ---
DATE : 11/01/2026 00:44:09
INPUT ANALYSÉ :
[Link: https://scirate.com/arxiv/2402.05072]

01. INDICE DE RÉALITÉ PHYSIQUE : 85%
Commentaire : Travail solide de physique statistique hors équilibre (matière active). L'approche hydrodynamique fluc

...(continued)

Oh..... Oh no.... the method starts by doing unnecessary heap allocated copies of all the inputs...

Found it here: https://github.com/tequilahub/pauliengine/blob/aa4f8259beb770169943802677294f92bb0e9f59/src/PauliString.h#L175

⭐⭐⭐⭐⭐

Where is the source code located? I wanted to compare your pauli multiplication subroutine to the one in Stim. Stim similarly uses bitwise operations (11 per word to be exact) and popcounts to perform A *= B including the phase. Stim's subroutine is shown in simple form in figure 2 of the paper http

...(continued)

Good to know! I'm not looking into 4.8.8 myself, but I have talked to another PhD student who is apparently working on this. Maybe their STIM infrastructure will also prove useful to implement simulations of some of your logical gates!

A very nice result, no doubt, but I would advise to look at [this PRL][1] from 20 years ago (which you cite) and revise the claim of the first sentence of the abstract.

[1]: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.96.043003

Thank you for your interest in our work, and sorry for not responding sooner.

We don't have immediate plans for simulating the planar 4.8.8 code, but we might consider doing so in the future if there is interest.

Dear Zhenyu and Shinsei,

thank you very much for your impressive arXiv:2601.00761 on improved Monte-Carlo sampling of Pauli distributions!

Here, I would like to point out our SciPost Phys. 19, 085 (2025) ([https://doi.org/10.21468/SciPostPhys.19.4.085][1]) where we show perfect pauli sampling for

...(continued)

Hey Craig, thanks for the comments!

1. What you said about separating the T gates and Cliffords, is exactly how we do it in Theorem 6 :) (and this also reduces the T-count to O(C+n) instead of the naive way which is O(C+nD))
2. And the XYZ=i trick is how Lemma 5 works! We just applied a fanout of t

...(continued)

Cool stuff. I like the controlled constructions using borrowed qubits to enable parallelism.

Here's an idea to reduce the depth overhead: separate controlling the Cliffords from controlling the T gates. Consider what would happen if you ONLY controlled the T gates but not the Cliffords. Then the

...(continued)

Ah, okay so it's only part of the circuit with full postselection. That removes the need for the tradeoff curve, because there's no freedom of choice. I now understand the point you are comparing to (the "ungrown" point in figure 1).

If we could go to higher distances... It would be interesting t

...(continued)

Hi Craig, thanks for the questions.

To clarify, we simulated only the injection and cultivation stages. We omitted the escape stage because it consists entirely of Clifford operations (which are well-understood) and requires CPU decoding that introduces overhead to our GPU workflow. Therefore, we

...(continued)

Very interesting. Nice to have some better ground truth on this.

Did you retain enough data from your simulations to make an equivalent of figure 14 from the cultivation paper, showing attempts-per-kept-shot vs error-rate-per-kept-shot? It's hard to do a comparison without knowing the *cost* of a

...(continued)

Hi Barbara, thank you for providing these references! We weren't previously aware of them, but they look very interesting / relevant to our own work. We will be certain to review them carefully.

Interesting work! Note that Section 5 in our paper https://arxiv.org/pdf/2303.13723 also notes that the toric rotor code is expected to undergo a KT transition (this was based on us writing down the corresponding stat. mech. model but no numerical investigations nor further published work), while we

...(continued)

What benefit is CI (or a variant thereof) supposed to bring for a large, weakly-correlated system like this, if e.g. DLPNO-CCSD(T) gives a near-exact answer? How will your SQD results stack against SHCI? (Esp. in light of Reinholdt et al https://arxiv.org/abs/2501.07231 ?)

I am puzzled by the proposed metric for Shor's algorithm in this work. It seems to impose no restrictions on the classical pre- and post-processing which opens up a whole can of worms, as I try to explain below (since the authors of this work explicitly invite dialogue).

1. Firstly, for problem ins

...(continued)

Nice paper! I noticed you also have schemes for the 4.8.8 code that generate the full Clifford group. Do you have any plans to do circuit-level simulations of these schemes? I'm quite interested in how the 4.8.8 circuit performs in general with the ancilla-free measurement circuit, because I think u

...(continued)

Just until this is addressed in v2: loglog(1/eps) depth is Thm 13.5 of the Kitaev-Shen-Vyalyi book. This is overall depth, not just T depth.

Thank you so much. I just realized this!

Dear Zhenhuan, if the group only contains the identity, the channel only needs to purify the maximally mixed state (the unique state in the algebra spanned by the group). It achieves this by always outputting the maximally entangled state (regardless of the input state).

Congratulations on the intersting result!

I was wondering the relationship between your result and Theorem 3 in [arXiv:2509.21111][1], which proves the exponential sample complexity lower bound of preparing a single purification state.

Your result seems to hold for all unitary groups. So if th

...(continued)

If you want to try it yourself Min's implementation of the beam search decoder is now available here: https://github.com/ionq-publications/BeamSearchDecoder

The term *light rectangle* was used [20 years ago by N. David Mermin with the same meaning](https://arxiv.org/pdf/gr-qc/0411069). Mermin also deduces the invariant interval from the area of a light rectangle drawn on the Euclidean plane.

Hi Ben, thanks a lot for your kind words!

Whether "optimal" should be reserved only for results that are tight without logarithmic factors is, we think, still somewhat up for debate. 🙂

For example, one of the two concurrent seminal papers establishing quantum state tomography optimal up to logs w

...(continued)

It's great as a quantum-inspired algorithm, but what are the prospects of ever executing this on FTQC, given the scaling of the sampling cost of VQE? I mean, isn't the number of samples required to reach a decent accuracy prohibitively high above ~10 qubits?

I fail to see a novel contribution in this pre-print. Since Shor put out his groundbreaking paper [[Shor94]](https://doi.org/10.1109/SFCS.1994.365700) [[Shor97]](https://doi.org/10.1137/S0097539795293172) back in 1994 many proposals for improvements have been published. Using divisors of the order $

...(continued)

This is an excellent piece of work, congratulations on the result! I am particularly excited to see query complexity bounds established for general quantum channels, going beyond states and unitaries.

I just wanted to clarify one point. Would it be more appropriate to use the term *near-optimal l

...(continued)

Congratulations on this nice paper! I have a small query regarding Algorithm 2 (Sliding-Window Decoding), specifically in lines 17 and 18. In the pseudo-code, m_com is passed to the inner DECODE function (line 17), while m_cvg is used to mask the error vector ê (line 18). However, based on the input

...(continued)

it might be interesting for you to know that somewhat related states were studied here from a different perspective (in particular using continuous groups)
https://arxiv.org/abs/2510.06602

Thanks for the reference!

Sadly, we were not aware of your work. The polygamous behaviour you describe is exactly of our interest. I'll look into that.

Congratulations on the results!

It is always great to see more work exploring these important properties of non-locality. I wanted to highlight our work from earlier this year studying bipartite monogamy/polygamy. PRR : https://journals.aps.org/prresearch/abstract/10.1103/1m7q-rgrh. (ArXiv version: https://arxiv.org/abs/2405.20286

...(continued)

Link to the repo: https://github.com/QoroQuantum/maestro

The work is interesting, but there are several major issues that make it difficult for me to fully trust the authors’ claims. Clarifying these points would greatly help my understanding.

1. It is unclear whether the experiments were actually conducted according to the definitions proposed in the

...(continued)

Without having looked at this paper, the following comments. The fact that one has to use a different norm for errors (diamond norm) for FTQC to cover general errors has been known since Kitaevs papers "Quantum error correction with imperfect gates" and "Quantum computing: algorithms and error corre

...(continued)

@Mingyu I suspect you made some mistake in the simulation, or were simulating a non-fault-tolerant construction (e.g. the output qubit being a single physical qubit). When I put 5-to-1 simulation into Quirk, I see the same qualitative "continuous perturbations become perturbations with pauses near t

...(continued)

Lovely stuff - just wanted to signpost our paper which also considers these errors at the resource state level, applying it to a full architecture scirate.com/arxiv/2507.16152. I'd say our work is most closely aligned with what you talk about, since we don't use polarization encoding and do consider

...(continued)

Interesting. I have some old simulation for 5-to-1 distillation that says otherwise (coherent rotations before the MSD circuit is suppressed but not those after it). Thanks for correcting, I might check this out

@Mingyu What? That's definitely false about the 15-to-1 distillation. One of the most common examples I show people when introducing them to Quirk is that if you put global continuous rotations at the end of the 15-to-1 distillation, and then look at the postselected output state, it rotates continu

...(continued)

Thanks, I'm glad that splitting-up helped!

Hi Craig,
Thanks for the question! Yes, it's not known if A implies B! The problem is not limited to fault-tolerance but also is there in error correction. So let's say A' refers to "a family of error correcting code having error correcting threshold for stochastic noise" and "not B'" refers to "

...(continued)

Right... maybe it's related to the difficulty of designing the entire FTQC architecture that works for both stochastic and coherent noise models, which this paper seems to do successfully. I can give you a really dumb example where typical protocols fail under the noise model where you have small an

...(continued)

But in the context of scaling-with-distance-or-not-with-some-threshold, that distinction shouldn't matter... right? I might need a bigger distance or a stricter threshold to ensure the bound holds despite quadratic growth rather than linear growth, but qualitatively that detail shouldn't be able to

...(continued)

I think the concern has been that in a stochastic error model, the error builds up linearly, but in a worst-case coherent error model where error is a sigma_z rotation by some angle theta, error builds up quadratically (p=sin^2(theta) then sin^2(n*theta) ~ n^2*p), so it has been hard to "translate"

...(continued)

I appreciate the splitting-up of 2408.05260 into two. It's a large paper with lots of results that could be missed. (At least, I missed some of them)

I have a background question: are there any known examples where a circuit is known to be fault tolerant to stochastic noise (call this "A") but not fault tolerant to general noise (call this "not B")?

I ask because I was surprised to learn it wasn't known that the A implies B. My intuition is that

...(continued)