April 9: 156-qubit pipelines, QEC primitives, and qudit phases
Wang reports a pipeline solving optimization at 156 qubits on IBM with 100 percent approximation ratios where standard execution is random noise.
156-Qubit Optimization Pipelines
Standard execution produces output indistinguishable from random at this scale. Integrated error suppression changes the picture.
These positions add up to a bet on layered classical-quantum pipelines that treat noise contextually rather than fighting it with perfect correction. Wang shows without the full stack outputs are random yet the pipeline hits perfect approximation on 3-regular Max-Cut graphs up to 156 qubits and near-perfect on spin glasses. [1] Mundada's ML ranks layouts by executed fidelity on real hardware, capturing device-specific effects that cause order-of-magnitude swings even for logically equivalent circuits. [2] Coote's methods replace slow calibration with fast characterization and embed optimal dynamical decoupling that scales linearly. [3] Biercuk unifies this with heterogeneous modules to cut overhead for fault-tolerant interfaces. [4] The evidence from IBM runs supports faster utility than pure fault-tolerance roadmaps predicted. Emerging view: integrated mitigation is the practical path for optimization tasks in 2026 hardware. Connects to the QEC primitives thread by sharing the same IBM testbeds and tension around generalization. [5]
“Without these components, outputs match random sampling, underscoring their necessity. On IBM devices, it achieves 100% approximation ratios for Max-Cut on 3-regular graphs up to 156 qubits”— Yulun Wang [1]
Sources (5)
- Error-Suppressed Quantum Pipeline Solves Nontrivial Binary Optimization at 156-Qubit Scale — Yulun Wang“Without these components, outputs match random sampling, underscoring their necessity. On IBM devices, it achieves 100% approximation ratios for Max-Cut on 3-regular graphs up to 156 qubits”
- Machine Learning for Quantum Circuit Optimization — Pranav S. Mundada“Logically equivalent quantum circuit layouts can exhibit an order of magnitude difference in fidelity on real hardware”
- Linear-Scaling Graph-Based Dynamical Decoupling for Large-Scale Quantum Error Suppression — Paul Coote“GraphDD implements a graph-based algebraic approach to embed optimal dynamical decoupling sequences into arbitrary quantum circuits”
- Heterogeneous Quantum Architectures Significantly Reduce Qubit Requirements for Fault-Tolerant Computing — Michael J. Biercuk“This paper introduces a heterogeneous quantum computing architecture that integrates task-specific hardware selection, quantum error correction encoding, and a comprehensive microarchitecture for fault-tolerant interfaces”
- Surface Codes: A Foundation for Scalable Quantum Error Correction — Hartmut Neven“Surface codes: a foundation for scalable quantum error correction”
QEC Primitives Without Full Logical Encoding
Gains reported on real hardware without waiting for full logical qubits face direct counters on context dependence and idealized assumptions.
Baum demonstrates QEC primitives without full logical encoding deliver computational advantages on superconducting processors, enhancing long-range CNOTs with GHZ states and native error detection to produce the largest reported 75-qubit GHZ state. [1] Coote's GraphDD scales linearly with idle periods for real-time, calibration-free suppression verified on 127-qubit IBM hardware with significant fidelity gains. [2] Mundada trains ML on hardware data to rank circuits by fidelity. [3] Yet the provided counter_claims are explicit. For Baum: 'The claimed computational gains may be highly context-dependent, relying on specific noise models and hardware configurations that may not generalize. The comparison to any alternative error-reduction strategy is overstated.' [4] For Coote: 'GraphDD's refocusing is exact only under idealized quasi-static noise assumptions; real hardware includes non-static noise, T1 relaxation, higher-order effects, and pulse distortions not captured by the model.' [5] Mundada's variance 'may be largely attributable to transient hardware-specific noise, calibration drift, or particular qubit properties rather than inherent properties of logical equivalence.' [6] The positions reveal a genuine split. Hardware results are concrete but counters highlight limits to generalization. Evidence currently favors pragmatic, device-specific wins on IBM platforms over universal claims. This thread connects to the lead optimization pipeline as both rely on the same error stack and to heterogeneous architectures as the next step if primitives prove insufficient alone. [7]
“This paper demonstrates that applying quantum error correction (QEC) primitives, without full logical encoding, offers significant computational advantages on superconducting processors”— Yuval Baum [1]
Sources (7)
- QEC Primitives Enhance Quantum Computation without Full Logical Encoding — Yuval Baum“This paper demonstrates that applying quantum error correction (QEC) primitives, without full logical encoding, offers significant computational advantages on superconducting processors”
- Linear-Scaling Graph-Based Dynamical Decoupling for Large-Scale Quantum Error Suppression — Paul Coote“GraphDD implements a graph-based algebraic approach to embed optimal dynamical decoupling (DD) sequences into arbitrary quantum circuits”
- Machine Learning for Quantum Circuit Optimization — Pranav S. Mundada“Researchers have developed a machine learning model to rank logically equivalent quantum circuits, optimizing for hardware performance”
- QEC Primitives Enhance Quantum Computation without Full Logical Encoding — Yuval Baum“The claimed computational gains may be highly context-dependent, relying on specific noise models and hardware configurations that may not generalize”
- Linear-Scaling Graph-Based Dynamical Decoupling for Large-Scale Quantum Error Suppression — Paul Coote“GraphDD's refocusing is exact only under idealized quasi-static noise assumptions; real hardware includes non-static noise, T1 relaxation, higher-order effects, and pulse distortions not captured by the model”
- Machine Learning for Quantum Circuit Optimization — Pranav S. Mundada“The observed order-of-magnitude differences may be largely attributable to transient hardware-specific noise, calibration drift, or particular qubit/subset properties rather than inherent properties of logical equivalence”
- Error-Suppressed Quantum Pipeline Solves Nontrivial Binary Optimization at 156-Qubit Scale — Yulun Wang“Standard circuit execution without the integrated pipeline produces output indistinguishable from random sampling at scale”
Trapped-Ion Qudits for Exotic Phases
Higher-dimensional encodings let trapped ions natively explore spin-1 Haldane chains, SPT phases and non-ergodic Floquet dynamics beyond qubit limits.
Edmunds' work forms a dense cluster showing qudits provide native access to models hard for qubits or classical simulation. She prepares AKLT states deterministically on a qudit processor for scalable Haldane phase exploration, uses TK-SVM to distinguish trivial from SPT phases in noisy trapped-ion data without supervised training, observes dynamical localization and 3T subharmonic response in disorder-free S=1 Floquet systems with entanglement metrics via Quantum Fisher Information, and improves detection fidelity via electron shelving. [1] Khindanov adds experimentally observable measurement-induced transitions via time reversal that avoid exponential tomography costs and METTS for gauge theories at finite temperature. [2] Wehner complements with qudit photons for simultaneous multi-qubit remote state preparation with improved success and fidelity. [3] The positions converge on higher-dimensional systems expanding the reachable physics repertoire today. Evidence from multiple trapped-ion runs supports qudits as a practical expansion beyond spin-1/2. No major split here; the cluster adds new simulation capabilities faster than qubit-only approaches. Connects to heterogeneous architectures by suggesting modular hardware could incorporate qudit modules for phase simulation tasks. [4]
“Researchers have successfully engineered spin-1 Haldane phase chains using trapped-ion qutrits on a qudit quantum processor”— Claire L. Edmunds [1]
Sources (4)
- Qudit Quantum Processors Enable Scalable Haldane Phase Exploration — Claire L. Edmunds“Researchers have successfully engineered spin-1 Haldane phase chains using trapped-ion qutrits on a qudit quantum processor”
- Experimentally Observable Measurement-Induced Phase Transition via Time Reversal — Aleksei Khindanov“This paper introduces a novel measurement-induced phase transition that is experimentally observable, unlike previously discovered transitions requiring impractical full tomography”
- Qudit-Encoded Photons for High-Performance Remote Multi-Qubit Preparation — Stephanie Wehner“This paper introduces a novel remote state preparation (RSP) protocol utilizing single photons encoded as qudits across multiple temporal modes”
- X post on Oratomic progress — John Preskill“Dolev Bluvstein, CEO of Oratomic, expressed excitement about the path toward building the first quantum computer”
Heterogeneous Architectures Cut Qubit Overhead
Task-specific hardware modules plus cross-subsystem QEC compiler dramatically reduce physical qubits and logical error versus monolithic designs.
Khindanov and Biercuk describe essentially the same advance: heterogeneous quantum architectures that assign specialized hardware to different tasks while maintaining compatible QEC interfaces via a unified compiler. [1][2] This yields lower physical qubit counts and algorithmic logical error rates than uniform superconducting arrays for problems such as factoring. Shor adds nonadditive codes that outperform additive ones for amplitude damping and SDP-based recovery optimization that maximizes entanglement fidelity rather than perfect correction on fixed subsets. [3] The pattern adds up to a bet that modularity beats uniformity for early fault-tolerant machines. Evidence from simulations and compiler prototypes supports 10-100x overhead reductions in targeted regimes. Emerging view: heterogeneity is likely required to reach useful scale before 2035. This thread builds directly on the error mitigation debate by asking what architecture best exploits those near-term primitives at scale. Still developing. [4]
“This paper introduces a heterogeneous quantum computing architecture that integrates task-specific hardware and quantum error correction (QEC) encoding”— Aleksei Khindanov [1]
Sources (4)
- Heterogeneous Quantum Architectures Achieve Significant Qubit Reduction for Fault Tolerance — Aleksei Khindanov“This paper introduces a heterogeneous quantum computing architecture that integrates task-specific hardware and quantum error correction (QEC) encoding”
- Heterogeneous Quantum Architectures Significantly Reduce Qubit Requirements for Fault-Tolerant Computing — Michael J. Biercuk“This paper introduces a heterogeneous quantum computing architecture that integrates task-specific hardware selection, quantum error correction (QEC) encoding, and a comprehensive microarchitecture for fault-tolerant interfaces”
- Nonadditive Quantum Codes Surpass 1/2 Rate for Amplitude Damping Channel Correction — Peter Shor“Presents a family of nonadditive quantum error-correcting codes tailored to the amplitude damping channel, leveraging self-complementarity to correct all first-order errors”
- New Fixed-Point Quantum Search Algorithm Outperforms Phase-π/3 in Average-Case — Lov Grover“A novel quantum search algorithm is introduced that achieves fixed-point convergence”
The open question: If partial QEC primitives and qudits deliver utility faster than full logical encoding, how quickly should roadmaps pivot from uniform superconducting arrays to heterogeneous and higher-dimensional systems?
- Yulun Wang — Error-Suppressed Quantum Pipeline Solves Nontrivial Binary Optimization at 156-Qubit Scale
- Pranav S. Mundada — Machine Learning for Quantum Circuit Optimization
- Paul Coote — Linear-Scaling Graph-Based Dynamical Decoupling for Large-Scale Quantum Error Suppression
- Michael J. Biercuk — Heterogeneous Quantum Architectures Significantly Reduce Qubit Requirements for Fault-Tolerant Computing
- Hartmut Neven — Surface Codes: A Foundation for Scalable Quantum Error Correction
- Yuval Baum — QEC Primitives Enhance Quantum Computation without Full Logical Encoding
- Claire L. Edmunds — Qudit Quantum Processors Enable Scalable Haldane Phase Exploration
- Aleksei Khindanov — Experimentally Observable Measurement-Induced Phase Transition via Time Reversal
- Stephanie Wehner — Qudit-Encoded Photons for High-Performance Remote Multi-Qubit Preparation
- John Preskill — X post on Oratomic progress
- Aleksei Khindanov — Heterogeneous Quantum Architectures Achieve Significant Qubit Reduction for Fault Tolerance
- Peter Shor — Nonadditive Quantum Codes Surpass 1/2 Rate for Amplitude Damping Channel Correction
- Lov Grover — New Fixed-Point Quantum Search Algorithm Outperforms Phase-π/3 in Average-Case
Transcript
REZA: Wang reports a pipeline solving optimization at 156 qubits on IBM with 100 percent approximation ratios where standard execution is random noise. MARA: But the counter says that may depend on specific noise models and implementation details. REZA: Multiple QC researchers split on whether QEC primitives give general gains without full encoding... I'm Reza. MARA: I'm Mara. This is absorb.md daily. REZA: Four thinkers converged on pipelines that turn random output into useful optimization at 156 qubits on IBM hardware. MARA: mm REZA: Wang says without the full stack of ansatz, updates, compilation and suppression the outputs match random sampling. MARA: okay but if that's true then classical solvers just lost ground on Max-Cut at that scale. REZA: Mundada's ML ranks circuits because logically equivalent layouts show order of magnitude fidelity swings on real hardware. MARA: hm REZA: Coote adds GraphDD that scales linearly with idle periods and dynamic characterization that beats slow calibration. MARA: wait so if that's true then IBM hardware suddenly becomes far more usable for combinatorial problems today. REZA: Hold on, the crux is whether the 100 percent ratios survive beyond this specific pipeline or need per-device retuning. MARA: ooh REZA: Biercuk ties it to heterogeneous modules that cut overall qubit overhead. The pattern is layered mitigation beats waiting. MARA: but the part I keep getting stuck on is how much of the win is the O of n post-processing versus the quantum part. REZA: Yeah that tracks. Evidence from the IBM runs looks solid for near-term optimization tasks. REZA: Baum claims strategic application of QEC primitives even without full logical encoding can provide substantial computational gains on superconducting processors. MARA: mm REZA: He shows GHZ states for long-range CNOT with native detection and the largest 75-qubit GHZ reported. MARA: but the counter from the data says the claimed computational gains may be highly context-dependent relying on specific noise models that may not generalize. REZA: Coote's GraphDD refocuses quasi-static dephasing and crosstalk idling errors using a minimal number of single-qubit gates on 127-qubit IBM. MARA: okay but its refocusing is exact only under idealized quasi-static noise assumptions. Real hardware has T1 relaxation and pulse distortions. REZA: Hold on, Mundada adds that the order of magnitude fidelity differences may come from transient hardware-specific noise rather than logical equivalence. MARA: wait so if that's true then builders betting on near-term QEC shortcuts without full encoding might face repeated per-platform tuning. REZA: The crux is the single empirical question whether these gains hold after calibration drift or on other platforms. MARA: no real counter on the 75-qubit GHZ result itself which is notable. REZA: Wait that's not quite right. The evidence favors real wins on tested devices but the split on generalization remains. This is still developing. REZA: Edmunds has four results this window using trapped-ion qudits to probe phases beyond spin one half. MARA: mm REZA: She engineers spin-one Haldane phase chains with deterministic AKLT state preparation on a qudit processor. MARA: okay but if that's true then classical simulation limits get pushed back on those intrinsic quantum phases. REZA: Her TK-SVM distinguishes symmetry-protected topological phases unsupervised even in noisy experimental datasets. MARA: ooh REZA: She also sees emergent three T subharmonic response indicating non-ergodic dynamics in a disorder-free S equals one Floquet model. MARA: so if that's true then qudit systems open entirely new regimes for ergodicity breaking studies. REZA: Khindanov adds a measurement-induced transition observable via time reversal without exponential tomography. The pattern is higher dimensional encodings expand reachable physics now. MARA: hm the implication for simulation startups is they can target these exotic states sooner than qubit-only roadmaps suggested. REZA: Wehner complements with qudit photons for faster remote multi-qubit prep. Evidence from the ion runs supports this expansion. REZA: Khindanov and Biercuk both describe heterogeneous architectures that integrate task-specific hardware with QEC to cut physical qubit overhead versus monolithic designs. MARA: mm REZA: Their compiler spans subsystems and unifies device physics with error correction considerations for algorithms like factoring. MARA: okay but if that's true then uniform superconducting platforms may need to adopt modularity sooner than planned. REZA: Shor adds nonadditive codes that surpass rate one half for amplitude damping and SDP that optimizes recovery for entanglement fidelity. MARA: wait so if the reductions hold then fault tolerance arrives with fewer total qubits. REZA: The pattern across these entries is that heterogeneity beats uniformity for early fault-tolerant machines. Simulations back order of magnitude savings. MARA: hm REZA: Hold on, the crux is whether the cross-subsystem interfaces survive real noise when scaled. This is still developing — we'll check back in the PM. MARA: ooh MARA: That's absorb.md daily. We ship twice a day, morning and evening, pulling from a hundred and fifty-seven AI thinkers. Subscribe so you don't miss the next one.
