Five Fault-Tolerance Preprints of 2 and 3 September 2026 Test the Assumptions Behind Low-Qubit Q-Day Estimates: Logical Gates on qLDPC Codes, Ion Crystals, GKP Lattices, Noisy Links and Decoders
A quantum error-correcting code is judged first by how well it keeps a logical qubit alive, and the past two years of results have been about that. The harder judgment is what the code lets a machine do while the qubit stays protected, because every estimate of how many physical qubits it takes to run a cryptographically relevant computation depends on the cost of the logical gates, and on high-rate codes that cost has been the open question. Between 2 and 3 September 2026 five groups posted preprints that each address a distinct component of that computation: the gates, the hardware to run them on, a different physical encoding, the links between modules, and the decoder.
The five are a paper by Rahul Sahay, David Long and Vedika Khemani on logical gates for quantum low-density parity-check codes, a proposal by Guo-Xian Tang, Lu-Ming Duan and Yu-Kai Wu for running such codes on a two-dimensional ion crystal, a construction of quantum low-density lattice codes by Timo Hillmann, Jens Eisert and Francesco Arzani, a study of distributed fault tolerance over noisy links by Moritz Schmidt and five co-authors, and a decoding method by Yuanqi Liu, Weilei Zeng, Junjie Wu and Lingling Lao. Two of them, Sahay's and Tang's, are about computing on a qLDPC code directly; the other three are about the encoding, the links between modules and the decoding that any such computation would need, and together they touch most of the quantities a resource estimate has to assume. This post reads what each one claims, sets the five against the Quantinuum result this site read two days ago, and follows the thread to the resource estimates that set the post-quantum calendar.
Practical takeaway. A physical-qubit estimate for a cryptographically relevant attack on an RSA or elliptic-curve key depends on several coupled quantities: the physical error rate and the target failure probability, the code and its overhead, the logical gate count of the algorithm and the cost of each gate on that code including the ancillas and distillation it needs, and the connectivity, parallelism and cycle time that turn all of it into a runtime. High-rate codes have cut the overhead. This week's papers work on the gate cost, the links and the decoder, and any estimate a committee is shown should say which gate constructions it assumes and whether they have been demonstrated on hardware.
What Sahay, Long and Khemani's Chain Map Hierarchy Adds on 2 September 2026
Quantum low-density parity-check codes encode many logical qubits in comparatively few physical ones, which is why they appear in every recent estimate that cuts qubit counts. Their weakness has been operations: on the surface code a small library of logical gates is well understood, while on high-rate codes each new gate has been a separate construction. Sahay, Long and Khemani's paper, "Computing with qLDPC Codes by Climbing the Chain Map Hierarchy", arXiv:2609.02999, posted 2 September 2026, sets out a single framework that treats logical Pauli, Clifford and non-Clifford operations in the same homological language already used to describe Pauli logicals. The authors define a family of chain complexes, the chain map hierarchy, whose homology classes encode logical unitaries and code surgery at any level of the Clifford hierarchy, so that the familiar trick of deforming a Pauli logical by a stabilizer becomes a search procedure for constant-depth implementations of Clifford and non-Clifford gates.
The paper's stated results are concrete. Using the search, the authors report constant-depth unitary implementations of the full logical Clifford group on any number of blocks of the two-dimensional toric code, including within a single block, and addressable logical CCZ gates on any triple of logical qubits across any number of blocks of the three-dimensional toric code. Beyond those codes, they report addressable non-Clifford gates on codes with many encoded qubits, and they show the hierarchy contains earlier constructions of logical gadgets, including a universal parameterization of cup products. What the abstract does not report is a hardware demonstration or a circuit-level error simulation; the contribution is a language in which new gates can be found. That places it on the theory rung of a readiness ladder, and it is the paper of the five most likely to change what the other four can build.
Four Companion Papers: an Ion Crystal, GKP Lattices, Noisy Bell Pairs and a Better Decoder
Tang, Duan and Wu's paper, "Frequency-Multiplexed Parallel Gates for Quantum LDPC Codes in a Two-Dimensional Ion Crystal", arXiv:2609.04081, posted 3 September, attacks the hardware side of the same problem. High-rate codes need nonlocal entangling gates for syndrome measurement, and the usual trapped-ion answer, physically shuttling ions, slows as the qubit count grows. Tang, Duan and Wu instead propose parallel nonlocal gates on a two-dimensional ion crystal through frequency multiplexing, with adiabatic conditions that suppress gate infidelity and crosstalk and hold up against slow drift in the trap frequency, which they identify as a leading ion-trap error source. Their numerical example is a [[248,10,18]] bivariate bicycle code on a crystal of 512 ions, ten logical qubits at distance eighteen, for which an optimized qubit mapping and frequency assignment reaches a logical error rate of 10−12 under what they describe as realistic noise parameters, with moderate laser power. It is a simulation of hardware that does not yet exist at that scale, and the reader should hold it as such.
Hillmann, Eisert and Arzani's paper, "Quantum low-density lattice codes", arXiv:2609.03021, posted 2 September, moves the low-density idea into a different physical setting and is about encoding, not about qLDPC computation as such. Gottesman-Kitaev-Preskill codes encode discrete information in the continuous states of a bosonic mode using a lattice, and native GKP codes have been hard to construct and decode. The authors co-design decoder and code by lifting classical low-density lattice codes into GKP families, report code properties comparable to or better than concatenated GKP-surface codes on the same number of modes, and show that a fully analog, linear-time message-passing decoder performs close to hybrid decoders; they promise the source code in two open Julia packages. The paper by Schmidt, Moureau, Rodatz, Poór, Mounzer and Grans-Samuelsson, "Resource-adaptive distributed fault tolerance with very noisy Bell pairs", arXiv:2609.03048, takes the question across modules, on the surface and color codes: when the only link between two chips is a shared Bell pair far noisier than an on-chip gate, how is a stabilizer measured across the seam. Their reported result is that integrated decoding halves the distillation code distance required, compared with distilling entanglement and decoding separately, so far fewer Bell pairs are consumed, with circuits that adapt to the Bell-pair rate or to the auxiliary qubits available, benchmarked on the surface and color codes under circuit-level noise with interconnect noise added.
The fifth paper, "Approximate maximum-likelihood decoding via truncated free energies" by Liu, Zeng, Wu and Lao, arXiv:2609.03928, posted 3 September, is about the classical computer beside the quantum one. Maximum-likelihood decoding gives the lowest logical error rate a code can reach under known Pauli noise, and exact evaluation is #P-hard, so practical pipelines fall back on minimum-weight decoding, which keeps the single lowest-weight correction and discards the rest. The authors recycle the candidate samples that stochastic decoders throw away into a per-class truncated free-energy estimator, provably bounded on both sides, and return the logical class that minimizes it at linear classical overhead. On toric and color codes they close up to 83 percent of the threshold gap between minimum-weight and maximum-likelihood decoding, with the largest single improvement on the 6.6.6 color code under depolarizing noise, from 17.28 to 18.62 percent. On the [[144,12,12]] bivariate bicycle code the method cuts the logical error rate by 13 percent at a physical error rate of 0.05 relative to minimum-weight decoding on the same candidate pool, with no matching enumeration and no code-specific tensor-network contraction.
Why Gates on High-Rate Codes Enter the Q-Day Arithmetic: the Oratomic-Led 10,000-Qubit Estimate of 30 March 2026
The reason five theory papers belong in a governance publication is a paper from March. On 30 March 2026 Madelyn Cain, Qian Xu, Robbie King and six co-authors including John Preskill, a team whose listed primary affiliation is the neutral-atom company Oratomic, posted "Shor's algorithm is possible with as few as 10,000 reconfigurable atomic qubits", arXiv:2603.28627, which states in its abstract that optimized estimates for cryptographically relevant instances had required millions of physical qubits and that the new figure follows from leveraging advances in high-rate quantum error-correcting codes, efficient logical instruction sets and circuit design. Under what the authors call plausible assumptions, a system of 26,000 physical qubits could run the discrete logarithm on the P-256 curve in a few days, with RSA-2048 one to two orders of magnitude longer. The three ingredients in that sentence are among the problems this week's papers work on: the codes, the logical instruction set, and the circuits that implement it, with the decoder and the inter-module links as the parts an estimate has to assume beside them.
That estimate is already cited in migration plans. Cloudflare cited it, among several developments, in its post-quantum roadmap of 7 April 2026, which targets completion of its own network migration in 2029; the gate cost examined here is one input to such an estimate and did not by itself produce that date. And the estimate is one of the two March results that open "Before Q-Day: The Race to Quantum First", the War on the Rocks essay of 20 July 2026 by Mauritz Kop, the founder of this site, and Joseph Federici of the US-China Economic and Security Review Commission. Their argument turns on lead times: cryptographic migrations historically take a decade or more, the slowest systems to change are the ones that matter most, and if the machine arrives in 2035 the migration still has to start now. The essay reads the resource estimates as premises for policy, which is the right posture, since each estimate rests on gate constructions of its own. What this week adds is a look at how those constructions are made.
How Quentir Reads It
Two days before these five papers, this site read Quantinuum's C4-Helix result on Helios, a twenty-ion code that computed the full Clifford group on two logical qubits and reported the error per logical Clifford beside the physical rate. That is the experimental end of the same question. Sahay, Long and Khemani's framework describes how to find such gates on much larger codes; Tang, Duan and Wu's proposal describes an ion machine that could run one with 512 ions instead of twenty; the decoding paper describes how to squeeze more from the classical side once the machine exists. The experiment measured two logical qubits at distance six; the proposal simulates ten logical qubits at distance eighteen, and the gap between those two numbers is the work still to be done. Bharti, Haug and Tanggara's proof, which this site read in August, that fault tolerance carries an unavoidable logarithmic overhead, sets the floor under all of it; the papers of this week are about how far above that floor a real gate set has to sit.
For a chief information security officer or a program office the reading is procedural. Every qubit-count estimate arriving in a briefing rests on an assumed logical instruction set, and this week shows that set is still being constructed for the codes the low estimates rely on. The question to put to any vendor or analyst who cites a number is which gate constructions it assumes, whether they are constant-depth or require distillation, and which have been run on hardware; an estimate that cannot answer is a research target with a date attached, which is still a reason to plan. The migration side of that plan starts, as this site's Defense Monitor read in the ASD and NCSC guidance that both begin with the cryptographic inventory, with knowing where the vulnerable keys are, and the theory does not change that order.
The items to watch carry their own dates: a circuit-level error simulation of the chain-map gates on a specific high-rate code, which Sahay, Long and Khemani leave for later work; a hardware demonstration of frequency-multiplexed gates on an ion crystal larger than a few dozen ions; and the next revision of a cryptanalytic resource estimate that names one of this week's constructions in its assumptions. Quentir's Quantum Defense Evidence Register is built to keep the published resource estimates, the codes they assume and the hardware results they rest on side by side on one readiness ladder, so a committee can see which numbers have a demonstrated gate set beneath them. The public analysis on this site continues daily at quentir.ai/blog.
Sources: Rahul Sahay, David M. Long and Vedika Khemani, "Computing with qLDPC Codes by Climbing the Chain Map Hierarchy", arXiv:2609.02999, 2 September 2026, for the chain map hierarchy, the constant-depth Clifford group on toric-code blocks, the addressable CCZ gates on the three-dimensional toric code, the non-Clifford gates on codes with many encoded qubits and the cup-product parameterization. G.-X. Tang, L.-M. Duan and Y.-K. Wu, "Frequency-Multiplexed Parallel Gates for Quantum LDPC Codes in a Two-Dimensional Ion Crystal", arXiv:2609.04081, 3 September 2026, for the frequency-multiplexed nonlocal gates, the trap-frequency drift, the [[248,10,18]] bivariate bicycle code on 512 ions and the 10−12 logical error rate. Timo Hillmann, Jens Eisert and Francesco Arzani, "Quantum low-density lattice codes", arXiv:2609.03021, 2 September 2026, for the GKP lattice construction, the comparison with concatenated GKP-surface codes, the analog linear-time decoder and the Julia packages. Moritz Schmidt, Martin Moureau, Benjamin Rodatz, Boldizsár Poór, Elie Mounzer and Linnea Grans-Samuelsson, "Resource-adaptive distributed fault tolerance with very noisy Bell pairs", arXiv:2609.03048, 2 September 2026, for the halved distillation distance under integrated decoding, the resource-adaptive circuits and the surface and color code benchmarks. Yuanqi Liu, Weilei Zeng, Junjie Wu and Lingling Lao, "Approximate maximum-likelihood decoding via truncated free energies", arXiv:2609.03928, 3 September 2026, for the #P-hardness of exact maximum-likelihood decoding, the truncated free-energy estimator, the 83 percent gap closure, the 17.28 to 18.62 percent threshold on the 6.6.6 color code and the 13 percent reduction on the [[144,12,12]] code. All five were read from their arXiv abstracts and metadata this run. Madelyn Cain, Qian Xu, Robbie King, Lewis R. B. Picard, Harry Levine, Manuel Endres, John Preskill, Hsin-Yuan Huang and Dolev Bluvstein, "Shor's algorithm is possible with as few as 10,000 reconfigurable atomic qubits", arXiv:2603.28627, 30 March 2026, with Oratomic as the authors' listed primary affiliation, for the 10,000-qubit figure, the 26,000-qubit P-256 runtime, the RSA-2048 comparison and the three ingredients named in the abstract. Cloudflare, "Cloudflare's post-quantum roadmap" by Bas Westerbaan, 7 April 2026, for the 2029 target and the reference to that estimate among the developments behind the roadmap. Mauritz Kop and Joseph Federici, "Before Q-Day: The Race to Quantum First", War on the Rocks, 20 July 2026, for the lead-time argument and the two March 2026 estimates it opens on; the first author is the founder of this site. The three earlier Quentir posts linked inline, on Quantinuum's C4-Helix result, on the Bharti, Haug and Tanggara overhead theorem and on the ASD and NCSC inventory guidance, are this site's own prior readings and carry their own source lists. The reading that a resource estimate is only as good as the logical gate set it assumes is Quentir's own.
Published intelligence, built to inform your own decisions. Published: September 4, 2026.